PRE2016 3 Groep20
Group members
Lotte Aerssen - 0892039
Wouter van den Bemd - 0948482
Lennard Kerkhoven - 0955882
Noud Schoenmakers - 0938197
Bjorn Walk - 0964797
Wouter Weekers - 0956095
Introduction
Delivery services in the Netherlands are well up to date concerning the latest technologies. When ordering a package, a track and trace code is given to track where the package is located and it will also provide a indication for when it will arrive. However, high speed delivery is still not realized, as most fast package delivery services provide a 24 hours service. There are three reasons that show it has to be changed. First the delivery times are too long, customers and small businesses rely on these services to receive their packages as soon as possible. However, because of dense traffic, daytime only delivery and unautomated transport this takes long. Also in the changing society of today people are not bound to one location anymore as they travel more and more miles every day. To solve this problem society needs a system that delivers their packages on-the-go. Finally the services only deliver in a given time frame, thus the customer needs to stay home if he or she wants to receive the package.
Goal
These reasons show that the delivery system is in need of innovation. In this report research will be done on autonomous delivery systems in the Netherlands. An analysis will be made for the main stakeholder PostNL. In the Netherlands there are several other delivery services that are in competition with PostNL. In order to move on with this competition it is necessary to deliver the packages and letters as efficiently and fast as possible, in order to meet the demands of the customers in the best way. PostNL has the biggest market share (50-60%) in this sector in the Netherlands [5] which is why it is picked as main stakeholder. An autonomous delivery system would allow for 24 hours a day service and an accurate planning as computer controlled systems can constantly measure their progress and make an approximation of the delivery times. This means that the time frame in which the package is delivered becomes much smaller, which is great for the satisfaction of the customer. Also the aim is to design the system in such a way that the delivery times become much shorter. Finally one of the goals is to allow the customer to pick a location and a time frame, which allows delivery on-the-go.
All these design criteria should be satisfied if possible while keeping the possibilities of adaption of the system realistic. This means that the price of adapting such a system should be acceptable. Also it is important that the system should be built around the society instead of the society around the system.
This report will first analyze the different available options. After picking the most favorable option, research will be conducted on whether it is possible to implement the system in PostNL and how much better the system can become than the current system at the three mentioned design criteria.
Approach
The designed system has to overcome many problems, as each option has its own technical or ethical obstacles. During this project, one option for autonomous delivery will be picked. The analysis of the different options should result in a recommendation or advice on which delivery system to use based on five criteria. These criteria will provide an overview of the advantages and disadvantages of the system, which allows selecting the most favorable option. The criteria have been listed below.
- Effectivity
- Reliability
- Safety and privacy
- Protection of the package
- Costs
After picking the most favorable option, the obstacles for this particular option will be explained in detail, however, it is not possible to overcome all the problems, therefore the focus will be on one of the problems. This problem will be analyzed and a solution will be provided. This will consist of a literature analysis paired with data from experiments. Also an error protocol will be developed where as much as possible unwanted scenarios will be covered. At the end of the report the solution will be provided for the selected problem together with an overview of the remaining problems for this method of delivery and finally a technical advice for PostNL which gives a possible solution for the design of the autonomous vehicle itself based on data that was gathered during the project. In order to fully realize the system, follow up research will need to be conducted on the remaining problems.
Autonomous delivery systems
In terms of vehicle choice there are several suitable options for package delivery with drones. These vehicle types can be divided in aerial and ground types. Each different vehicle has different advantages and disadvantages and these will be described in this project in order to conclude witch vehicle is the best, according to the standards in this project.
Influence of population density
Urban and country delivery systems need a different approach when it comes to autonomous package delivery. Urban zones are high populated areas which contain many buildings which can be tall. Urban zones are usually located in cities or towns and contain many roads. Country areas are very different to urban zones because they have a low population density with fewer but longer roads of which some may be unpaved. High buildings are less likely in country areas but trees are more common.
Delivery system for rural areas
Delivery systems in rural areas preferably have a high travel speed as great distances are common. High speed wheeled robots that travel on the road and flying robots meet these requirements. Flying robots are preferable since they do not need advanced hardware to deal with other traffic. Another downside of wheeled robots is that the technology to take place in traffic is also not fully developed. Flying robots can travel easily over the often low buildings and trees in rural areas with a very low chance of having a dangerous crash with a person.
Delivery system for urban areas
In urban areas, a way to travel is preferred where the complexity of the environment does not affect the robot too much. Safety is also an issue, since it is a highly populated area. Slow moving robots are preferred because this makes collisions less dangerous. Legged robots or wheeled robots which travel on the sidewalk are perfect for this environment. Low travel speeds do not matter since travel distances are usually small in urban areas. The sidewalk is a relatively simple environment, which makes movement easier. Flying robots and high speed wheeled robots are not safe enough because the complexity on the road or in the air between the tall buildings makes it likely to crash, which is rather dangerous for such high speed systems.
USE aspects
In the proces of an autonomous delivery system there are also several other actors that are involved. Here all the actors are described and there role in the final step of the process, the landing, as well.
User
- The main users are the customers of the delivery company. They use the service of the autonomous delivery system to get what they want, their ordered product. These users are mainly individuals that can be anywhere, so not necessarily at home. The users may be concerned about if the drone can land close enough to them, if not someone else can intercept their order of what happens when they end up being unable to receive their package while the drone already has arrived.
Society
- The people in traffic are an important part in the society that can get affected by drones. Drones fly through the air so during their traveling they may only get in tough whit planes, but their landing places should not be close to where people may walk around.
- Neighbors of the customers may be affected by drones that fly and land close to their houses.
- The authorities may be anxious about the kind of products the drones are delivering. Are they able to intercept drones to check if they are transporting illegal content such as drugs of illegal fireworks.
Enterprise
- The main enterprise is the delivery company. In this research the focus lies on PostNL. They use the drones to fulfill their customers needs. They might be concerned about safe landing places for their drones. The done has to be unharmed by for example the environment. Also they need a large amount of options for drone landing places in order to make it attractive for their customers to use their services.
- Specific web shops might be interested to have products delivered at a lot of locations. For example shops who sell comping gear that might be broken during a customer’s trip, of perhaps food or medicine. Delivering products on every location might make some products more attractive to buy
Options for delivery
There are many different methods to move from point A to point B. One obvious method is by using wheels. Robots on wheels are well developed but they require pavement to function reliably and they cannot handle stairs. Travel speeds of wheeled robots are usually slow if they travel on the sidewalk where high speed motorized vehicles are not allowed. Driving on the road allows higher speeds, but the robot has to be equipped with more advanced hardware, thus raising the price. Another way to move is by using propellers to fly, which can be very fast. Important disadvantages are that packages cannot be very heavy, tall buildings have to be avoided and a crash in a populated area can be dangerous. Finally legs can be used to move from one location to another. At the moment legs are one of the most challenging ways to move around as a robot, since balancing on different surfaces can be quite a challenge. This way of movement is similar to the robots on wheels. The main difference is that a robot with legs can walk stairs and is less likely to get stuck on unpaved roads.
Helicopter
The helicopter is an airborne vehicle which can carry a package. This particular option has a lot in common with the system used by Amazon Prime air. This small and lightweight vehicle type can only carry relatively light loads, as weight has a strong impact on battery life. It usually travels very fast at a certain altitude. In case of Amazon Prime Air this height will be 100 meters and the speed will be up to 80 km/h. The drone will carry the package in a safe, which can only be opened by the authorized person using his or her smartphone.
Effectivity
As helicopters are not affected by the quality or presence of roads they can be used in both urban and rural areas, making them very diverse and suited for all kind of environments. Package delivery requires a place to land, which renders it useless at locations without a proper landing spot. However, the spots do not have to be big, as the vehicle itself is small. The helicopter is also very effective at places with a bad infrastructure, because it can fly over obstacles. They also travel very fast and always travel the shortest route, which gives the fastest delivery time besides the plane compared to the other options. Maximum package size and weight is somewhat limited, since a helicopter has to constantly fight gravity and increasing the weight will increase the load. The range of a helicopter is also limited, as flying uses a lot of battery power and big battery packs can not be used due to weight.
Reliability
Helicopters are heavily dependent on the weather. Fast winds, pouring rain and snow all affect the helicopter. Of course there are modifications available to make a helicopter waterproof or increase the strength of the motor, but this also increases the weight and therefore decreases flight time and increases the power consumption of a helicopter.
Safety and privacy
Overall, the chance of a collision and a person or property, is very small for a helicopter. This is because it travels in the open air, with almost no obstacles to collide with. However, in the event of a crash, a helicopter will drop from a high altitude, which could be a very dangerous situation. This can be prevented by applying several backup systems in such a helicopter, so it can for example still fly with one or two malfunctioning motors. On the subject of privacy, the drone will need cameras and sensors to land at a certain position. This could violate privacy, however, if the images are automatically interpreted and deleted by the helicopter afterwards, it will not influence privacy at all.
Protection of the package
The fact that this option is airborne means that it has less risk of being damaged assuming there are traffic rules for drones. Its altitude ensures that the system can not be abused by a third party. The only moment where it is vulnerable is at the delivery point. Therefore it should not stay long at the delivery point, even though it has a protected safe.
Costs
The price of such a system is heavily dependent on the assumptions. Making a helicopter water resistant or adding more safety precautions may increase the price of the system.
Plane
A package delivering plane will generally travel very fast, as this is required to generate enough lift for the plane to carry the package. This vehicle can only carry lightweight packages, as weight has a strong impact on the battery life. This vehicle will travel at an altitude of about 100 meters, which is the same altitude as Amazon Prime Air. Typical speeds for radio controlled aircraft are between 120 and 180 km/h. Dropping the packages will require parachutes and a certain drop off location.
Effectivity
This option is definitely the option with the fastest delivery time. With the highest travel speed of all options and the fact that it can fly straight to the target, the plane will be able to deliver a package within minutes. The maximum package size is somewhat limited, as extra weight makes the flight time lower. The fact that it ignores waters, roads and other obstacles, makes the plane useful in areas that cannot always easily be reached.
Reliability
Planes are dependent on the weather. Fast winds, pouring rain and snow all affect the helicopter. Of course there are modifications available to make a plane waterproof or increase the strength of the motor, but this also increases the weight and therefore the flight time decreases and the power consumption increases.
Safety and privacy
Airborne vehicles have only small chances of collisions in air, which can be avoided with air traffic laws. One problem with planes is that they cannot stop flying when needed. Helicopters for example can slow down and then slowly decrease its altitude until it touches the ground. Planes need a runway to stop, which might become a problem above urban areas. Crashing of a plane can be severe due to the high speed and altitude at which it travels. Planes will also need multiple sensor to calculate how to drop off a package. This may offend privacy if the images are not immediately deleted.
Protection of the package
A plane will need to land on a landing strip if it wants to deliver. Another way might be delivering a package which requires dropping it from the plane. Packages will have a parachute attached to it and they will be packed in protective wraps, which will ensure a safe landing for the package. A problem with this is that there is no good way to prevent unauthorized persons to grab the package, as the person cannot authenticate with his or her mobile phone while the plane is midair.
Costs
The costs of such a plane is heavily dependent, adding modifications can cost a lot of money but may increase reliability. One side effect of the plane is that protective packaging materials will be used, which will increase the costs per delivery.
Walking robot
This robot uses legs to move on the sidewalk from point A to point B. It uses sensors to avoid collisions with obstacles or persons and it can use GPS service to move itself towards the destination. This type of robot is very similar to the already built Cassie robot. Using legs instead of wheels allow the robot to walk on stairs and decreases the chance to get stuck on bumps on the sidewalk compared to wheels. The movement speed of the robot will be about the movement speed of an average pedestrian.
Effectivity
Delivering packages at the speed of an average pedestrian takes a very long time. Therefore the robot is only effective in a small range around the package disposal station. For covering large areas, a lot of disposal stations have to be placed and these also have to be supplied with goods. This is fine for small systems where only one disposal station is required, for example a pizza delivery service. The products will then be produced and disposed at the same location. This removes the struggle of filling the right disposal stations with the right products which makes the process much more slow because of the extra step in the transport process and therefore less effective. A system like this would only work well for small areas with a dense population. It can be concluded that the system will be effective in urban areas for local companies and less effective or even obsolete for greater delivery areas or a rural area with a low population density.
Reliability
Reliability is an important factor for a delivery system. Small delivery robots can be very reliable under the right conditions which leads to a trustworthy delivery system. Walking robots are unlikely to damage the package. This is because they move very slow and store the packages in a safe inside the robot, which in the uncommon event of a collision prevents damage to the package. In an event of hardware failure, the robot will also be rather safe, as the robot moves very slowly. The central station will be notified since the robot does not move. A repair mechanic can be send towards the robot to fix it, which causes a slight delay in delivery time but no damage to the package.
Safety and privacy
A walking robot only participates in a few ways in traffic. First of all it must be able to safely walk on the sidewalk using its sensors to avoid obstacles and humans or pets. A collision with a person will be very unlikely because distance sensors can check whether there is an object in from of them and then automatically stop moving. Therefore the only real threat is a human or vehicle which bumps into the robot, which can be more dangerous. Also, the robot must be able to cross a road, which is more complex than walking on the sidewalk, as it has to recognize whether it is safe to cross, taking different traffic into account and their different roles and speeds. This can become quite challenging but autonomous cars by Google have proven that autonomous driving is possible. This also includes an extensive monitoring system which can determine where every traffic participant around the vehicle is located and which speed it has.
Protection of the system and the package
A walking robot will be a very vulnerable robot, as it moves slowly and it is not that heavy due to its maximum size. The robot will need some sort of package protection safe which can be opened using NFC on the smartphone of the customer. To prevent stealing the vehicle as a whole, anti-theft features should be implemented. This can be done by using the tracker that is already in the robot, or by using a loud alarm which goes off once the robot is lifted up or damaged. This should also immediately send a warning message to the staff of the delivery system who can warn the police if it is necessary.
Costs
The costs of such a vehicle might be high, as it needs many sensors to observe its environment. Out of all these sensors the visual sensors are the most expensive as it needs both the cameras and computational power to process the images and locate all the obstacles. Also, the fact that many disposal stations will be required makes this system even more expensive.
Sidewalk cart
This vehicle moves packages from point A to point B using the sidewalk. It has various sensors to avoid collisions with obstacles or persons and it is able to use GPS to move itself towards the destination using wheels. The packages these robots can carry can be quite heavy, however, the volume of a packet will be limited. This is because the robot has to be able to avoid obstacles on the sidewalk, which cannot be done if the dimensions of the cart including the package are too big. The movement speed of the robot will be about the movement speed of an average pedestrian. This has been tested already in Washington DC where the robot delivers food to peoples doorsteps[18].
Effectivity
A small delivery unit that travels at the speed of an average pedestrian which is about 5 kilometers an hour is very slow. This makes the robot only effective in a small range around the package disposal station. For covering large areas, a lot of disposal stations have to be placed and these disposal stations also have to be supplied with goods. This is fine for small systems where only one disposal station is required, for example a pizza delivery service. The products will then be produced and disposed at the same location. This removes the struggle of filling the right disposal stations with the right products which makes the process much more slow because of the extra step in the transport process and therefore less effective. A system like this would only work well for small areas with a dense population. There can be concluded that the system will be effective in urban areas for local companies and less effective or even obsolete for greater delivery areas or a rural area with a low population density.
Reliability
Reliability is an important factor for a delivery system. Small delivery robots can be very reliable under the right conditions which leads to a trustworthy delivery system. Sidewalk carts are very unlikely to damage the package. This is because it moves very slow and stores the packages in a safe inside the cart, which in the uncommon event of an collision prevents damage to the package. The robot however has a chance to get stuck by for example driving on uneven terrain. The robot will be able to notice that it is stuck and send a warning to the central station, where a human controller can try to unstuck the robot. In an event of hardware failure, the robot will also be rather safe, as the robot moves very slowly. The central station will be notified since the robot does not move. A repair mechanic can be send towards the robot to fix it, which causes a slight delay in delivery time but no damage to the package
Safety in traffic
A sidewalk cart only participates in a few ways in traffic. First of all it must be able to safely walk on the sidewalk using its sensors to avoid obstacles and humans or pets. A collision with a person will be very unlikely because distance sensors can check whether there is an object in from of them and then automatically stop moving. Therefore the only real threat is a human or vehicle which bumps into the cart, which can be more dangerous. Also, the robot must be able to cross a road, which is more complex than walking on the sidewalk, as it has to recognize whether it is safe to cross, taking different traffic into account and their different roles and speeds. This can become quite challenging but autonomous cars by Google have proven that autonomous driving is possible. This also includes an extensive monitoring system which can determine where every traffic participant around the vehicle is located and which speed it has. The question however is how much a sidewalk cart can monitor as it is such a small vehicle.
Protection of system and the package
Stealing packages, destroying the equipment or stealing the robot as a whole are all forms of abuse. A sidewalk cart will be a very vulnerable vehicle, as it moves slowly and it is not that heavy due to its maximum size. The robot will need some sort of package protection safe which can be opened using NFC on the smartphone of the customer. To prevent stealing the vehicle as a whole, anti-theft features should be implemented. This can be done by using the tracker that is already in the robot, or by using a loud alarm which goes off once the robot is lifted up or damaged. This should also immediately send a warning message to the staff of the delivery system who can warn the police if it is necessary.
Costs
The costs of such a vehicle might be high, as it needs many sensors to observe its environment. Out of all these sensors the visual sensor are the most expensive as it needs both the cameras and computational power to process the images and locate all the obstacles. Also, the fact that many disposal stations will be required makes this system even more expensive.
Road cart
This vehicle delivers packages by traveling over the normal roads cars use. Because of this the road carts are much like the delivery vans that are used today. However there are a few differences to make it an autonomous delivery system. Firstly it drives completely autonomous. To make this possible it has to have a lot of sensors, GPS and AI to make decisions. Secondly the unloading of the package is also done autonomously. Due to its large storage area the road cart can carry large and heavy items. The road cart can travel at the same speeds as normal delivery vans can.
Effectivity
As said before the road cart is basically like the vans used today. This makes them effective in a large range, because they can travel at high speeds. Also the large storage space the vans have makes it possible to take a lot of packages in one run. This however introduces another difficulty namely the one of taking the right package at the right location. However there are already systems that can do this, making the road cart an option for both rural and urban areas.
Reliability
To be reliable the delivery vehicle has to be able to deliver the package safely and on the scheduled time at its destination. Road carts travel by the roads between the normal traffic. This makes its arrival times uncertain as there are a lot of things that can delay the vehicle, such as traffic jams or crashes. Another problem is possible damage to the packages. Because the storage room of the vehicle is large there is a lot of room for the package to move around when taking turns or crashing if it is not tightened secure enough causing possible damage to it. Also, when having either a software or a hardware failure the autonomous cart can become a dangerous projectile as it can travel at high speeds making safety an issue.
Safety in traffic
The road cart participates in the busy traffic on the roads. Therefore it has to have sensors to prevent collision. This technology is already used by for example Tesla for the autopilot function, so this should not be a problem. The only problem is thus when another vehicle crashes into the cart, but this is a problem all vehicles on the road face so there is no reason this should be the reason not to choose this option compared to the vans used today.
Protection of the system and the package
A road cart will not be very vulnerable to abuse, because it will most likely be the size of a normal delivery vehicle so it cannot be picked up like for example the sidewalk cart. The road cart can also be locked automatically when no package is getting dropped off. So abuse of the road cart will not be a big problem.
Costs
The costs of the road cart might be very high as it needs a lot of technology to make sure the package gets delivered safely, but with the rapid growth of technology these techniques will get cheaper.
Zeppelin
One of the options is to make the system by using a zeppelin. It will take the packages from A to B through the air. It will use GPS to know its location and use some sort of engine to propel itself through the sky.
Effectivity
A zeppelin can easily be adjusted to have a high lift force. Using a large balloon with a lightweight gas e.g. helium makes it possible to lift heavy objects. It could therefore be used to easily lift packages of 2 kilograms, while they might even be able to lift up to 5 kilograms or more. This is a big difference with other aerial options. Also the power usage is much lower on a zeppelin, as the balloon passively compensates for the gravity caused by the package. The delivery speed of such a system is rather low as zeppelins move slowly through the air, which is mainly due to their large frontal area which causes a high air resistance. As an aerial vehicle it is able to fly directly to the destination, which is an advantage over ground vehicles. However, its size can make landing difficult.
Reliability
As zeppelins are large lightweight flying vehicles, they are easily disturbed by the weather. Differences in temperature can cause the gasses in the balloon to expand or compress, which will affect the lifting performance of the vehicle. Another issue is wind due to the large size of the vehicle. Winds can easily sweep away a zeppelin, as the large surface catches much wind and its low mass makes it extra vulnerable.
Safety in traffic
When a zeppelins balloon is pierced and the gas flows out of the balloon, it drifts towards the ground on a slowly. Therefore, in an event of hardware failure, the possible damage of such a system would be much less compared to other aerial vehicles. However, due to its size it could become a big obstacle. If for instance a zeppelin lands on a highway, the balloon could be very dangerous because it could cover the view of a driver, which leads to a crash. Also several sensors will be required to percept its environment. One important sensor is a camera to allow it to detect where to land. This may violate privacy, as the camera may be able to record private property.
Protection of the system and the package
As the zeppelin flies in the air, the package is relatively safe. Due to their slow speed and soft landings in case of hardware failure, it is very unlikely that the vehicle itself will damage the package. Also a safe will be added which prevents unauthorized persons from receiving the package from the zeppelin.
Costs
The costs of a zeppelin will be around the same as the other aerial vehicles. While the motors and the battery can be smaller and cheaper, the balloon increases the price. This is mainly due to refilling with helium due to leakage and the extra maintenance costs as the balloon should be checked on leaks, which is more time consuming then replacing a motor when its malfunctioning.
Favourable option selection
Effectivity
The walking robot travels at the speed of a pedestrian. In this way, delivery takes a very long time, certainly in larger areas. The side walk cart is not much faster. Looking at the other options, the walking robot and the side cart are not what we are aiming for. However, these systems might work very good in small urban areas with a dense population. The road cart is similar to the current system, which we want to improve. They are effective in both rural and urban areas and there are already systems available for doing this autonomous. The helicopter is a very effective option, because it is capable of flying over obstacles and it is not using roads. The helicopter is suited for all kinds of environments and they are very fast. However, the package weight might be a big limitation. A plane would be a faster way to deliver packages, the packages could be heavier than in a helicopter, and like the helicopter, the plane ignores roads and other obstacles. One major disadvantage of using a plane, however, is that it needs a landing strip. The last option is the zeppelin, which is a very safe way to deliver packages. It doesn’t crash like helicopters and planes do. In addition, it can carry very heavy packages and consumes very little power.
Reliability
The zeppelin might seem as a very nice option, but it is a very slow system. On top of that, it is a very big object, which can be easily hit by other objects. Also landing is complex and the light weight of it is a drawback, because it is very dependent on the weather. The sidewalk cart, on the other hand, is very reliable, because there is little chance that the package will be damaged. However, when the robot gets stuck, it will not be capable of getting free by itself, which will cost a lot of time. The same holds for the walking robot; the package is safe, but if something happens, it will take a lot of time. Speed and accuracy are in favor of a plane or a helicopter, but planes and helicopters are very dependent on the weather. There are solutions for this, but this will impact the flight time and the consumption of the systems.
Safety and privacy
The chance of getting into a collision is not very high for a helicopter or a plane, but when they do, the damage probably is big, and might lead to dangerous situations. However, there are several backup systems possible, so that the helicopter and plane will still be able to fly or land safely in case of one or two malfunctions. For the walking robot, a collision is also very unlikely. The only problem might be people or vehicles bumping into the robot and crossing roads. The same holds for the side walk cart. For both autonomous systems, however, there are programs to safely cross the road. These programs can also be implemented in the road cart. For instance, Tesla and Google car drive autonomously with such a program. The zeppelin, as described before, is a very safe way of travelling. A big drawback is that the vehicle is very big, so it can be hit by objects very quickly. Protection of the system and the package The zeppelin might be a safe way of traveling, but the system can be easily abused. Very little things can cause the zeppelin to crash. This is not the case in the other options. Safety of the package is another important aspect that should be considered. Because of the lifting force, it is easy to lock the package in a safe in the zeppelin. The road cart, the walking robot and the sidewalk cart can also be locked pretty easily. However, the walking robot and the sidewalk cart can be picked up and stolen fairly easily as well. So there should be a solution that prevents that from happening. This is not the case with the plane and the helicopter, but they, just like all options, need some kind of system for authentication. How does the system know who grabs the right package? NFC is probably a good solution for authentication for people using a smartphone or tablet. This is, however, a problem for people not using a smartphone or tablet.
Costs
The costs for all options depend on assumptions, should it be water resistant? How many safety precautions should be added? How many and what kind of sensors should be used? The plane, however, needs extra protective packaging materials, which will increase the costs per delivery. For the walking robot and the sidewalk cart, many disposal stations will be required, which will increase the costs for these options. The costs for the road cart may also be somewhat higher, because of the technology needed to safely deliver packages and to make sure the right package is delivered at the right house.
The choice
We chose the helicopter as the delivery system that is most likely to be implemented in real life. The helicopter is compact, it is capable of landing almost everywhere. Compared to the other options, it is not necessarily more evironmentally friendly. However, it is at least more environmentally friendly than the delivery system right now, with vans and cars. On top of that, it is certainly the most efficient option. The air is rarely used except for airplanes, but those fly much higher than helicopters. Therefore, we think helicopters are an exceptional new way of delivering packages to regular customers. In addition, they are the second most fast option. Only the plane is faster, but there are a lot of drawbacks in using a plane in a delivery system. To finish, it is relatively cheap. The initial costs are quite high, however, further on, there will be almost no additional costs.
Stakeholder background
Delivery process
The package delivery process has these following steps:
1. The client makes an order on a web shop
2. The package gets send from the web shops stockroom to a sorting centrum
3. The package is processed in the sorting centrum
4. The package is send to a pickup point close to the clients home
5. The client picks the package up from the pickup point
PostNL has 6 sorting centers in the Netherlands [6]. From these centers the packages are transported to the pickup points with large trucks. If the customer demands that the package is delivered to their house directly, the package is transported with a smaller van directly from the sorting center. Because of the low number of sorting center these vans usually have to traverse long distances and transport a lot less packages. In this process PostNL delivers 500.000 packages a day, during heyday this can increase to 1 million [7]. The pickup points in step 4 and 5 of the process are spread across the country, but in the more populated areas there are a lot more pickup points than in the smaller villages. In 2015 there were around 6,8 million people in the Netherlands that live in the big densely populated cities and around 6,5 million people on the rural eras. The rest of the population lived in a transition area between urban and rural [8]. The people that live in the urban areas do not have to look far for a pickup point and because the pickup points can provide for many customers in the neighborhood this is perhaps the most efficient way. But especially for the rural population the pickup points have to sustain customers spread over a large surface. Because there are still a significant amount of people living in these areas it is worth to improve the services in these areas.
Right now if customers choose to have their package delivered to their front door, the vans of the delivery company have to send a van with a very limited package capacity to an area where several customers expect an order. The van has to take the most optimal route to deliver the packages. The points in this process that an autonomous delivery system can improve are:
- costs of delivery employer
- costs of fuel for the van
- speed of delivery
- efficiency of delivery
Price classes
The way the shipping cost of a package is determined depends on the weight of the package. Usually there is also a maximum size of length and width that the package needs to have. Lastly the shipping cost can be increased if it is requested that the package gets send with urgency. The price classes, based on weight, of PostNL for example are divided like this: [2 kg], [2-5 kg], [5-10 kg], [10-20 kg], [20-30 kg].[2]
Number of packages
It is important to know how many packages PostNL delivers a day to discuss if the use of drones will be realistic or not. The next table is formed out of the data given by PostNL itself.
![Delivery table PostNL [10]](/wiki/File:PostNL_table.png)
As can be concluded out of this table, the mail in the Netherlands is getting less over the years but the parcels are getting more. If we take 2012 as reference year and thus take it as 100 %, the following table is calculated with using the following formula: Percentage in comparison with year 2012 = ((Nieuw-Oud))/Oud*100%
![Delivery table PostNL[10]](/wiki/File:PostNL_percentagetable.png)
And to make it more clear that the delivery of parcels is getting higher and higher while the Mail in the Netherlands is getting lower, the next table is formed in which the percentages are calculated against the previous years.
![Delivery table PostNL [10]](/wiki/File:PostNL_previoustable.png)
As can been seen in the table the delivery of mail in the Netherlands is drastically going down while the delivery of parcels is going up with almost 50 percent in 2016 in comparison of 2012. Because the amount of parcels a day is necessary for determining, the average parcels per day in 2016 is calculated out of the data received out of tables. There are about 365.242 days in a year and because PostNL does not deliver on Sundays there will be 365.242 – (365.242/7) = 313 days left.
This means that there are averaged 177x10^6/313 = 565379 parcels a day delivered. This number should be a bit lower because of low delivery weeks and holiday weeks such as Christmas or Sinterklaas where a lot of parcels are ordered. Also business packages are included in this number and it goes up to every weight. Because we only want to look to private orders and parcels with a maximum weight of 2 kilogram this needs to be calculated for this average value.
However PostNL does not want to give us any information about how the delivery of packages and their weight is divided and also if it is for business or private. And looking to other companies will not give the right values at all.
Legislation
The Netherlands
There are no clear rules for autonomous drones stated in the legislation of the Netherlands, therefore the rules for professional use will be used as rules for the autonomous drones. In the legislation drones are divided in light drones which weight up until max 4 kilogram takeoff weight, and heavy drones which weight up until 250 kilogram takeoff weight. For packaging drones the focus will be on the light drones. The drone always need to give priority to all other air- and land-traffic when they are approaching. They also need to fly in daylight and may not fly in the dark. The drones may not fly higher than 120 meter and need to have a minimum distance from crowds, compact building density, artworks, harbors, industrial areas, railways, public roads, vessels and vehicles of 50 meters. There are also No-Fly-Zones in the Netherlands were no drones are allowed to fly. As you can see are the No-Fly-Zones mainly in big cities and airports. The red area's are the the No-Fly-Zones for the drones.
The European Union
The rules and regolation about drones in the European union are given by the European Aviation Safety Agency, also called EASA. Although they state clear rules about the use of "Unmanned aircraft" (drones), the paper is still a roadmap or prototype of how it should be in the EU. EASA made a risk based approach to regulation of unmanned aircraft in the so called "JARUS concept of operation" and in this chapter the import conclusions will be stated. The JARUS concept of operation is developed for two main goals. The first goal is integration and acceptance of drones into the existing aviation system in a safe and proportionate manner. And the second goal is fostering an innovative and competitive european drone industry, creating new employement, in particular for small and medium sized enterprises (SME's).
The concept of operation should be regulated in a manner proportionate to the risk of the specific operation. This means that the broad range of operations and types of drones should be taken into consideration and it is proposed to establish three categories of operations and their associated regulatory regime. these three categories are the proposed regulatory framework of the JARUS-paper. The three categories are Open, Specific and Certified. Below the catergories are introduced and shortly explained.
The open category
This category contains of very low risk drone operations, this is why they do not need involvement of the Aviation Authorities, even for commercial operations. Therefore no air-worthiness approval is foreseen and there are also no licenses or approvals needed for the operator and/or pilot. The drones are designed to fulfill simple operations and for small and medium-sized enterprises to gain experience. The risk for other airspace users is very small and moderated through separation with “manned” aviation. The drone may only fly when the following rules are satisfied:
• Under direct visual line of sight (VLOS): 500 m
• At an attitude not exceeding 150 meter above the ground or water
• Not around specified reserved areas (such as airports, environmental and security)
The risk for people on the ground is very low because through the use of low battery/energy aircraft and setting a minimum distance with respect to people. Also flights above crowds are prohibited, but flights above the people which are not related to the operation in cities or populated areas is allowed. And because there is no airworthiness approval required, industry standards could be applied for making these drones.
The specific category
This category should cover the operations that do not meet the specifications of the open category where a certain risk needs to be mitigated by addition operational limitation or higher capability of the involved equipment and persons. For this category the operator should perform a safety risk assessment, identifying mitigation measures, that will be reviewed and approved by the National Aviation Authority. Therefore the operation should be analysed on each specific aviation risk and an authorization is necessary.
The certified category
This category is used when aviation risks are similar to normal manned aviation. These operations and the aircrafts involved therein would be treated in the classic aviation manner. Multiple certificates would be issued as for manned aviation plus some more specific to unmanned aviation. The difference between specific operations and certified operations is very small but could be based on kinetic energy considerations, type of operations and the complexity of the drones notable in terms of autonomy.
Because the idea is that the drones will be fully autonomous, they will most likely be positioned in the Certified category which means that the drones and their operations would be treated as the classic aviation manner with multiple certificates. The rules in the European Union are far from complete, which makes it impossible to make an reliable conclusion on whether its possible to use drones for as delivery option.
The United States
In U.S.A. the FAA already made rules for drones for commercial use. It will be most likely that the rules they have stated are going to be a lot like the rules the European Union will make in the future. Therefore the 10 most important rules for commercial use of drones in the U.S.A. and in particular delivery drones, are written down below.
1. Unmanned aircraft must totally weigh less than 25 kilogram, even if it is carrying a package.
2. Daylight-only operations, or civil twilight (30 minutes before official sunrise to 30 minutes after official sunset, local time) with appropriate anti-collision lighting.
3. May use visual observer (VO) but not required.
4. Must yield right of way to other aircraft.
5. First-person view camera cannot satisfy “see-and-avoid” requirement but can be used as long as requirement is satisfied in other ways.
6. Maximum groundspeed of 161 km/h (87 knots).
7. No carriage of hazardous materials.
8. Maximum altitude of 120 meters above ground level (AGL) or, if higher than 120 meters AGL, remain within 120 meters of a structure.
9. External load operations are allowed if the object being carried by the unmanned aircraft is securely attached and does not adversely affect the flight characteristics or controllability of the aircraft.
10. Requires preflight inspection by the remote pilot in command.
These rules are directly copied from the FFA, Source: https://www.faa.gov/uas/media/Part_107_Summary.pdf
Main sources: http://spectrum.ieee.org/automaton/robotics/drones/faa-announces-commercial-drone-rules http://spectrum.ieee.org/aerospace/aviation/a-parallel-air-traffic-control-system-will-let-delivery-drones-fly-safely https://www.faa.gov/news/press_releases/news_story.cfm?newsId=20515
source: https://www.dronewatch.nl/wp-content/uploads/2016/06/EASA_EU-Roadmap-Operation-Centric-Approach-Drone-Ops-v13_160620.pdf source: http://www.agoria.be/nl/Europa-eensgezind-over-regels-voor-Drones source: https://www.easa.europa.eu/easa-and-you/civil-drones-rpas source: https://www.easa.europa.eu/system/files/dfu/UAS%20Prototype%20Regulation%20final.pdf https://www.easa.europa.eu/system/files/dfu/204696_EASA_concept_drone_brochure_web.pdf
Prognosis
The scenario in this project assumes the perspective of the Dutch company PostNL operating in the Netherlands. But in the previously in this section it was found out that there are no active rules in the Netherlands, the European Union is only having rules in development, clear rules for the scenario in this project have to be taken else were. The American rules of the FAA are very suitable because they are mend for commercial usage of drones specifically, which is the case in this scenario. Also as mentioned before these rules could be taken over by the European Union, because they also see the usage of drones as a big opportunity and can be inspired by these already defined rules, especially if they prove to be successful in the United States in the future. Therefore the rules of the FFA are assumed as the rules that the end result has to take into account. This means that the final concept of a package delivery drone should operate while following these rules.
The concept
In this paragraph, the concept for the autonomous delivery system is given. This includes the interface that customers experience, the steps that need to be taken to deliver a package, the problems that the system may cause and an overview of the actions the drone has to perform and what to do in case something goes wrong.
Delivering packages with a drone comes with some limitations. There is no way to deliver piano’s, matrasses or other big packages with a drone. They are too big and/or too heavy. Therefore, we stated some specific limitations with respect to size and weight. The maximum size of a package is 40x40x40cm, and the maximum weight of a package is 2kg.
Immersive problem
Specified aspect
After describing the focus that this investigation will be on, an aspect of the total problem is chosen to describe. This regards the specification of the landing cycle of the delivery drone. This aspect brings up safety of the user, reliability of the drone, technological problems regarding the localization and many more. Therefore, a fresh look is needed to overcome these problems. In this chapter, first the problems are described and categorized. This brings up the discussion of the importance and possible solutions regarding the problems. Finally, all solutions are brought together to generate a complete idea of the delivery cycle and a conclusion is given to this subject.
General Problems
The landing cycle includes many problems. Let’s start to make an sunny day case on the landing cycle. Ideally, the drone flies to the location based on the GPS information that has been given with the order. However, this information might not be 100% correct, because there is a 5m range, in which the drone is not at the right house. The drone corrects for this inaccuracy and assures its position and starts to descend. When it has come out of the ‘airspace’ and it floats on ground level, the drone specifies its destination. In the mean time, the drone informs the user that the package will be delivered in several minutes. After doing so the drone lands and waits for the user to grab the package and, when necessary, signs for the package. The sunny day case is described in a flowchart.
Fail plans
To specify the problems that arise while programming the drone can be described by questioning every move the drone makes. The assurance of its position, for instance, could go wrong when connection fails. When this happens there are a few things that it could do. One of these things is that the drone could give an error and return to the postal service or postal truck.
In addition, the drone could have problems connecting to the user. When this problem arises the drone picks a safe location close to the address it has to be. Then it floats above street level so no one is annoyed by the zooming object. Then, the drone sends a message to the user, telling its location. When the person does not respond the drone flies back to the postal service. However, when the drone connects to the user it waits and responds when the customer has arrived. This gives a specific situation where the error occurs that the drone can’t find the user. This gives a difference between going to the user, such as in the sunny day case, or the user coming to the drone when it fails to connect.
The last problem regarding the sunny day case is how to verify the user. This could be done using the QR code on the landing pad. Also the user will have to be verified, therefore, the user could have another QR code for the specific package. This gives an advantage to the current system, where the customer does not have to give a specific identification.
Additional problems
There are however more problems with the whole drone delivery system than only the landing part. But because all the problems should be investigated separately because they stand on its own, they are not fully explained in this paper. However in this paragraph a small introduction is given about how we envision the other main problems and what is suggested as advice to those problems.
Privacy
The first problem which is encountered by using drones for delivery is privacy. Because the drone works autonomously, an actual camera is not needed for him to know where to fly to. However because it is advised as rule by the EASA and FAA to obligate that every drone should have a direct visual line of sight, the drone will need a camera on which the observer can see where it is flying if it is necessary to interfere (for example when there is a system failure). Therefore PostNL have all the videos of where the drones are flying and can harm the privacy of people when for example flying over someone’s garden.
Therefore the advice to PostNL is that they should not store those videos for longer than a day or even do not save it at all if possible and make it clear to the society that the videos that will be stored are save, private and will only last for, for example, one day and will not be used for any other kind of operations than taking over drones when there are issues with it. They should make it clear that even the police or other parties cannot use the videos for investigation and such. When all this is made very clear, for example by commercials and explanations on the site) to the whole society it will probably be more accepted and the drone will be trusted as a delivery drone and not as a spy gadget used by the government or anything like that.
Safety
This will be a major problem when using drones as delivery systems because there are so many factor what can go wrong. The drone can still be seen as a small helicopter and should be handled with care and therefore reckless behavior of people can have a major impact on the acceptance of the drones. Also weather can have impact on the safety of the drones, when for example a drone falls on the ground because of strong wind.
Therefore the advice to PostNL is that they should make strict and sound rules about how people should handle these drones and talk with the government about giving certain fines and even prison sentences when some rules are discarded by a certain person. And because weather conditions are also a big part of the system, the advice is to have close contact with the weather stations or even get a small department where the weather is closely followed and advice is given if they should stay on the ground or can fly. Because if they may not fly, PostNL can still use the normal way of delivery.
Distribution centers
PostNL have only six distribution centers in the Netherlands and for drones to fly to the center and pick up the packages is just not an realistic option. therefore the use of drones from this point is not a good alternative for the current system because the drones need to fly an way bigger distance than the drones battery can hold. Also if the drone will fly a longer distance, more things can go wrong and the risk factor is increased.
Therefore the advice to PostNL is that they maybe should find a middle way if they do not want more distribution centers and small distribution centers. The middle way is driving a truck with the drones and packages to a small distance, for example a distance of a community, and then use the autonomous drones for the delivery of the parcels. Therefore the basis and time of delivery is improvement because more drones can be used at the same moment and no time is spent on driving through busy streets or waiting for people to open there doors.
Drop off at location
By choosing the helicopter option instead of for example the road cart another issue was introduced: finding a landing place. Because where road carts can only stop in front of a house and drop a package there the helicopter can land practically everywhere. So we will have to think about ways of finding this location to land (if landing is even necessary) to drop of the package. Below a few options are listed:
GPS
The most well-known location finding technologies nowadays has to be the Globally Positioning System or GPS for short. To guide the helicopter from the distribution center to near the package delivery destination GPS will be used. But when it comes to landing there could be other options as well. If we look at using GPS to find the landing position there is namely a bump in the road. GPS is only accurate to around 5m when under open sky, and performs worse when there is something over it like for example a bridge. This inaccuracy is not a big problem when the drop-off location is in the middle of a big meadow, but when delivering in the middle of cities 5m of inaccuracy could mean landing on the street. It speaks for itself that this is not safe for the helicopter itself and the person retrieving the package. This kind of problem could be solved by using some sort of fixed landing spot.
Landing pad
The first sort of such fixed landing spot is something Amazon will use when Amazon Prime Air becomes reality. They want every customer to have a portable landing pad at home on which the Amazon logo is printed. The Prime Air drones will scan the area for this pad when they are at the approximate location and then land on top of it. This solves the problem of the inaccuracy of the GPS as the helicopter will have a fixed landing spot to focus on. A problem with this system is that the pad has to be visible from up in the air. Another problem is that when your neighbor has also ordered something the helicopters do not know which one has to land on which pad.This last problem could be fixed by embedding some sort of unique code in each landing pad which the helicopters can recognize.
Beacon light
A variation of the landing pad to fix the neighbor problem could be a beacon light. This beacon will flash with a fixed frequency which the helicopter will recognize. The helicopter can then land on top of this small beacon in the same way as it would land on the landing pad. The problem which such a beacon is again that is might not be visible from the air. On top of that there could be the problem that due to daylight the helicopter will not see the beacon flash.
Drawn landing space
Another option would be to make it possible for people to draw their own landing area on for example the street with chalk or removable paint. This could be done by for instance drawing a large dot surrounded by two concentric circles. The helicopter then has to search for this drawing using its cameras and approach it in the same way as the landing pad or light beacon. The problem with this idea is that none of the drawn landing spaces will look exactly the same, so the helicopter will have to have very advanced pattern recognition abilities and a universal landing place drawing has to be agreed upon. On top of that rain could wipe out the drawing leaving the drone without a landing spot.
Waiting person
Instead of using something to land on the helicopter could also just look for a person that is waiting for a package and land near him or her. The problem with this is that there is no way how the helicopter can recognize who is standing in the street waiting for their package and who is waiting for something else, especially in busy streets. On top of that there are problems with how close a drone can come to a human before it gets uncomfortable (see also the wiki PRE2015 3 Groep 2). This causes the need for a lot of free space around the waiting person, which might not be present in the busy streets of a city.
Hovering
Apart from landing at the right location using one of the methods above or a combination of them it is also possible to hover at a specific height. The person retrieving the package could then detach the package from the helicopter without it having to land. This would come in handy in areas where landing might be hard such as the very south of The Netherlands where there are hills so the roads are not 100 percent horizontal. Another place where this could come in handy is with multiple story buildings, where the package could be delivered at the right floor level instead of having to go all the way down to the main entrance. Unfortunately we will have the same problem as mentioned in the previous paragraph, namely that people tend to get uncomfortable when a drone approaches them too closely. If the helicopter cannot get within an arm’s length distance from the retrieving person without him or her getting uncomfortable the option of hovering would not work.
Conclusion
From the above it is clear that for the helicopter to determine where to land it has to at least use GPS. However, as this is not accurate enough to deliver the package at the exact wanted location another option for recognizing the landing location has to be added. A few of the above options, namely 'waiting person' and 'hovering', are not viable because as said before people do not like drones to come near. This leaves three options which narrow down to the same idea: a visible marking at the location of delivery. The 'drawn landing space' has the problem that most people will not be able to draw the image accurately enough for the drone to easily recognize it. On top of that the drawing could be wiped out by rainy weather conditions. Therefore this option is also closed out, which only leaves the landing pad and the beacon light. From these two the landing pad is the best option as it does not have problems when sunlight reflects from things which could be mistaken for the beacon light. The chosen option is therefore the landing pad on which an unique (QR-)code will be printed so the drone can distinguish between your landing pad and your neighbors' pads.
Customer interface
Right now customers of PostNL who want to send a package can make a decision between two services. A regular service and a high speed one. High speed package delivery is able to deliver a package the next day before 10 am, while regular package delivery will deliver it the next day during work hours. In the new system, the customer has three delivery options. The two current available options will remain and the third option, the autonomous delivery, will be added. For the third option, owning a smartphone which contains the PostNL delivery app is required together with a foldable delivery landmark.
There are two ways to implement the system. This is because sending a package to a person will be different from a person ordering a certain package. The delivery will be the same, however, the system does not know whether the receiver is eligible to receive a package due to the fact that not everybody has a smartphone containing the PostNL app. The first way to solve this problem is to limit packages sent by this system to only apply to customers who order packages for themselves by for example buying a product on a web shop. This solves the information problem as the receiver knows whether he or she owns a smartphone with the app installed. The second way is to solve this problem is by using a system where customers have to sign up for drone delivery first. This allows PostNL to disable the option when the receiver is not registered.
Features of the PostNL app
The PostNL app will be available in recognized app stores as Googles Play Store and Apples App Store. After downloading the app, registration will be required. This is mainly required to check whether the owner of the app really lives at his or her address. After registration, PostNL will deliver a test package, which contains a foldable delivery landmark which contains a QR-code that contains a personal code for each customer, a registration code and a guide on how to use autonomous delivery. This landmark. In order to finish registration, the code should be entered on the smartphone. The activation code should be different from the QR-code that is on the landmark, as the code on the landmark could be scanned by a third party.
The PostNL app has two main purposes, which is to allow persons to pick a delivery location and time for their pending packages and to authenticate with the drone. In the main menu, several buttons will have to be included. The required buttons have been listed below.
![PostNL app menu [10]](/wiki/File:PostNLAppMenuTable.PNG)
In the Pending Packages menu a list will have to be included which can show all the packages that have been marked as autonomous delivery. The customer can select packages from the list which gives a small menu. In this menu the customer can set a delivery location, either by selecting a location on the map or by manually selecting it by using the GPS sensor inside a smartphone. Also a delivery time has to be selected. When the delivery time is near, a notification will be given which shows that the landmark should be placed at the drop off location. This notification will also be stored in the notifications menu. This notification will show the last prediction for the delivery time of the package and it shows that the smartphone is required to authenticate. In the Track & Trace the prediction of the delivery time will be shown for each package. This can also include packages delivered without drone delivery, as PostNL already owns this information. After the drone lands, it will ask for authentication with the smartphone. After selecting Authentication in the main menu of the PostNL app, it will show an image of how to place it on the drone which will open up the package safe.
By implementing the system in this way, PostNL will only store data about whether a customer is registered for autonomous delivery or not. No additional personal information will be required. Also customers will not be forced to transfer, as the old delivery options still remain.
Logistics and drone actions
In this chapter the big picture of automated delivery using drones is described. The package belt to customer is elaborated. This gives an idea on the situation where it gives an idea on the density on the system.
General plan
To start of it is important to first note that the model will not immediately replace the current system. It will generally, quietly, overtake the system while it is imbedded into society. To do so the postal service will start using a small army of drones on certain locations in the Netherlands. The decision could be made to first take the uban areas or start with the rural areas. This is a question that remains unanswered. The fact stays that the business will start small and expand over time. The most important aspect of the plan is the location where the drones will pick up the packages. This question has many answeres since some companies plan to let drones an postal busses work together to deliver. This plan has the bus driving to a general location where all the packages in the neighborhood are deliverd by the drone that flies óut of the postal bus.
In this situation however the drones will directly fly to the distribution center. Making the delivery by drones much faster and direct. On the advantages and disadvantages a new study could be conducted, however in this report this way is chosen.
New concept of delivery
After clicking the payment button on a website to order a package, the system comes in to play. The first tasks of the distribution center remain the same, aka setting up the packages, labeling. After the packages are boxed the drone comes in, grabs the package and flies off. When left it immediately flies into the high, noncommercial, airspace (+/- 120 m). The drone flies to the GPS location that is given by the costumer. Afterwards the delivery is set in progress, this delivery cycle is described later in the concept chapter.
The system will start of as an extra option for delivery. Thus the package is still delivered in the ‘normal’ way, by postal truck. The extra option will be a drone option and will, in the beginning, cost extra. This is in the same way as the “fast delivery option” and the “late night delivery”. Therefor it will have a try out period. The customers that want to order by drone first have to be enlisted as a “drone-costumer” this means that they have means of landing for the drone, these means are elaborated later in chapter “Drop off at location”. After testing period where all found bottlenecks have been rectified. The postal service can choose to expand the business and give more priority to drone delivery systems.
Failure recovery
In the process of landing there are always things that can go wrong. For each of these cases, which we will call failures, a solution has to be found. Below the failures and their solutions are described.
Hardware failure
The first failure that can be encountered is the case where the hardware of the helicopter drone fails. An example of such a failure could be that one of the propellers stops working. Because of this the drone will lose part of its lift, which could cause it to start descending. When such a problem arises the drone will try to look for a safe spot to make an emergency landing using its on-board sensors (see also the Sensors chapter), this landing will make it vulnerable to vandalism but this will be discussed in the next paragraph. While doing the emergency landing the drone will send an error message to the postal service so someone can come and collect the drone. To keep the delay because of the failure to a minimum the following will happen depending on the location of the landing. If the destination of the package is near the location of the emergency landing the person collecting the drone will deliver the package himself by car, just as it is done nowadays. In case the emergency landing happens at a location where it would be quicker to take the drone to the distribution center and load the package onto another drone this will be done.
Vandalism
During most part of the delivery cycle the helicopter will fly at an altitude at which human interaction will be highly unlikely. However when landing it is possible that a person will vandalize the drone. This could be that the person damages the drone, tries to steal the packages or that it takes the drone and moves it. First the damaging will be discussed. In case the person damages the drone the protocol will be as mentioned in the previous paragraph: it will send a message and someone will come and collect the drone. To discourage damaging the drone it could be fitted with a camera which snaps pictures of its surrounding when it suffers damage so the vandal can be identified. The second way of vandalism could be when someone tries to take the attached package. This package is located in a box in the middle of the drone, which is locked with an NFC lock (see also PRE2 Groep1). When the drone notices that someone tries to force open the lock it could again take a picture. On top of that the drone could sound an alarm as to notify bystanders that someone tries to steal the package, much like a car alarm works. Lastly there is the stealing of the whole helicopter. As the drone is fitted with GPS it is possible to track the location of the drone, this makes it possible to always find the drone when it gets stolen. To let the postal service know when a drone gets stolen it will send again a message to the postal service when it is moved for more than 20 meters without doing anything on its own. On top of that the alarm could sound again. The radius of 20 meters is chosen as to rule out changing of the GPS location because of the inaccuracy of GPS and to allow for people to take the drone to a dry location to unload the package in case it rains, so they do not have to stand in the rain while unloading.
Extreme weather
The rain in the previous paragraph brings us to the point of extreme weather. Things such as strong winds, heavy rain or much snowfall could cause the helicopter to crash and thus damage the package. To prevent this from happening the first thing that will be done is to not send out drones when there is extreme weather in which a drone could not fly safely. In case the drone was sent out with weather in which the drone can fly safely and the weather turns extreme the drone will notice this because it will not stay on the track that it wants to follow. When this happens the drone will again try to land in a safe location as it would do with hardware failure. If then the weather gets better the drone can depart again and fulfill its task. This both indeed causes a longer delivery time, but it is a lot safer than drones crashing everywhere and the package will eventually be delivered. Moreover it is much the same like with the current way of delivering, because when weather is extreme now packages will not be delivered either.
Collision
Another failure that could occur is a collision between for example the helicopter and a bird or a building. When this happens there are two possible outcomes: either the drone suffers no damage and can continue its course or it suffers damage and cannot continue. In the first situation there is obviously no problem and the helicopter will deliver the package as it was supposed to. In the second situation the procedure will be the same as the hardware failure problem. When a collision happens the drone will always send a message to the postal service in which it says that it had a collision, at what GPS location, at which time and if it can continue its delivery. The postal service then knows where to look if there is damage done to other properties and when needed solve this with the owner of the property,
No 4G connection / connection failure
To be able to give an accurate estimation of its arrival time to the user the helicopter will have to have a 4G connection to send updates continuously to the application on the phone of the customer. When the drone loses its connection to the 4G connection it cannot send these updates to the application anymore. The drone will of course keep trying to connect to the network while it travels further to its destination. Meanwhile the application will still show an estimation of the time to arrival. This will be done in the same way as modern day navigation systems estimate the location of a driver when it drives through a tunnel: by taking its last known speed, location and heading and using that data to estimate where the drone is now and how much time is needed before it arrives at its destination. The biggest problem with the loss of the 4G connection is that the customer cannot cancel or delay its delivery in real time anymore. The drone will therefore continue its journey and when it arrives at the location where the package has to be delivered. If during this time the helicopter can connect again the information about if and when the package should be delivered will be updated and the drone will act accordingly. In case the drone cannot reconnect it will arrive at its location and not find a landing pad. What will be done in this situation will be described in the next paragraph.
No landing pad/ landing pad could not be found
When the drone arrives at its approximate destination it will look for the customer’s personal landing pad to land there. In some cases it is possible that the landing pad is not present, for example when the customer has cancelled its delivery and thus has not put his or her landing pad outside for the drone to land. It could also be possible that the landing pad could not be found because it is too far of the GPS location entered, because it is underneath a roof or tree, or because visibility is poor due to for example fog. When this happens and the drone cannot find the landing pad to land it will fly back to the distribution center and send an error message to both the postal service and the customer with the notice that the delivery has failed because it could not find the landing pad. The postal service can then check if the delivery was cancelled by the customer or if they need to send the package again. The customer then knows it has to place the landing pad somewhere else for a next delivery so the helicopter will be able to find it.
Wrong landing pad
When the drone arrives it searches for the right code on the landing pad. The problem could occur that it scans a wrong landing pad, for instance the customer its neighbors landing pad. This problem is solved by the drone, it simply keeps on searching for the next landing pad and if none is found the problem is dealt with as described above.
No authentication
After the helicopter has landed the customer will be able to collect its package from the drone. To verify that the person collecting the package is indeed the customer some sort of authentication will be used. As already said in the paragraph about vandalism one option is a NFC lock as mentioned in PRE2 Groep1. When the authentication fails or no authentication happens the compartment containing the package will not open. If after 30 minutes no authentication has happened and the drone has not been opened it will start to fly back to the distribution center and send an error message to the postal service describing the reason of its failure. The time-frame of 30 minutes is chosen because it would not be reasonable at all to leave the drone standing on the landing pad all day waiting for someone to collect the package. However the waiting time should also not be too short as the customer has to be able to get to the drone to collect the package.
No costumer
When the drone arrives at its location and zooms down to the landing pad, the situation could occur that there is no one that comes to the device to authenticate his/her package(s). When this situation occurs the drone is supposed to wait for 15 minutes before taking of and flying to the postal service. In the mean time the drone sends a message to the costumer as a reminder to pick up the package.
Unwanted or wrong location
In the case that for some unknown reason the helicopter would find itself in a place where it should not be, thus being at the wrong GPS coordinates, the drone should get back on track. To do this the drone will first look if from its current position it can reach its destination to deliver its package given its battery charge left and an estimation of how much battery charge it costs. If it can reach the destination it will go there. Should there be not enough battery charge left the helicopter will look if it can fly back to the distribution center, again taking into account the battery charge and the expected consumption. Finally if this is not possible the drone will try and land in a safe location and send a message to the postal service so someone can come and collect the drone. The person from the postal service will then either deliver the package himself or take the drone back to the distribution center to deliver the package with another drone, as mentioned in the hardware failure paragraph.
Additional recovery
If the drone has some problems and the recovery procedures, which are stated above, for certain problems did not succeed, then the final recovery is that the drone will be taken over by a operator. This is not only needed for the additional recovery, but it is also stated in the rules of the FAA that every drone can be taken over by the main controller. Therefore when there are no options left, the main drone operators of PostNL will need to take over the drone and fly the drone to safety. In order to know when the drone needs to be taken over or not, the drone has to give a signal to the operators of PostNL when they are in trouble and cannot recover from it. The operator will look at the signal and starts taking over the drone while getting to know the problem out of the given data of the signal. When the drone is taken over, the operator will fly the drone back to the distribution center or will let someone of PostNL know where the drone is landed or crashed. Taking over the drone will be done in the same way as the US army does with the Predator drone. The pilots however will have to go through extensive training. Such as the RPAS education in the Netherlands [20]. This will not come cheap since an average skilled pilot costs around €3000.-[20]. However, due to small chance of failure it is not expected that the drones will have to be taken over with high frequency. Therefor the stand-by drone recovery team does not have to be big. The best way of making the system as durable as possible is including the weather into the workforce at PostNL. For instance, when bad weather occurs the team has to be bigger compared to a sunny day. This can be implied into a model that gives the minimal amount of workers that have to be stand-by.
QR-code distance experiment
Introduction
This measurement will determine the maximum distance to reliably scan a QR-code. This includes testing different sized QR-codes, different cameras and different apps to scan a code. It is also important whether the QR-code can be detected while presented under an angle. The goal of this experiment is to research what distance between a drone and a landing pad is, for scanning a QR-code on the pad with a camera on the drone. Both vertical and diagonal distances are meaningful to test out.
Adding a QR-code to the landing pad creates a point of recognition for the drone. This makes detecting the landing location easy and personal landing pads can be used as well. This way the drone will not land on the wrong pad while delivering a package.
QR-codes consists of black and white pixels, the amount of pixels and therefore the amount of data is usually expressed in modules. The amount of modules can vary depending on how many information should be stored in the code. QR-codes use ASCII, this means that every stored character in the code can take on 256 different states. Right now there are 16,8 million inhabitants in the Netherlands. In the table below is listed how many different personal codes can be made with a number of characters.
As can be seen in the table above, three characters will provide enough personal numbers for every household in the Netherlands. However it is smart to pick four numbers as a system with three characters may not be sufficient in the future.
The most common used QR-codes use 25 or more modules, as this allows storage of URLS and other more useful information. A 25 module QR-code can store up to 26 characters of data. This is more than sufficient data for the personal code, as it can . QR-codes with less modules have relatively bigger black and white squares and are expected to be better readable for a camera. Therefore it is expected that the lowest amount of modules possible should be used.
One problem with this is that the smallest QR-code generator which is available on the internet generates a QR-code with 21 modules. This allows to store 9 characters of data, which is more than sufficient as it can provide up to 4*10^62 personal numbers. A code with the lowest amount of modules possible can thus not be generated by the generators currently available. Therefore a new generator should be made.
A test has been performed in 2011 on the minimum size of a QR-code. This experiment tested the maximum distance of reading a QR-code of a certain size. The conclusion of the experiment was that a QR-code with less modules is indeed better readable at longer distances and also a bigger image of the QR-code also improved the maximum range almost linear. During the test a 3-4 megapixel camera was used to detect a 25 module QR-code. It could scan a 46 millimeter wide QR-code at a distance of 45 centimeters, a 30 millimeter wide QR-code at 30 centimeters and a 15 millimeter wide QR-code at 15 centimeters.
When scaling these values, a 1 meter carpet would allow a scan distance of 10 meters. This can however be improved by better software, a better camera and a QR-code with less modules.[11]
Method
In order to find the maximum scanning distance, several devices will be needed. In the experiment multiple smartphone cameras will be used to test this maximum distance. Some of the important required devices are listed below.
Measurement environment:
On the wall there is a place to hang a QR code. In front of this code the tape-measure is then rolled out. With each phone is started close to the code. When the code can be read the person holding the phone moves back. This is repeated until the QR code can just bare be read anymore. The distance is then noted and the experiment is repeated with another sized code. To determine the influence of pixel density the experiment is done in almost the same way, except instead of using different sizes of code different pixel densities are used. Lastly the maximum angle at which a code can be scanned is determined by putting another tape measure parallel to the wall at different distances from the wall. Walking in the horizontal direction then gives a maximum horizontal distance, which can be converted to an angle using the distance to the wall.
Results
First of all it was found that the maximum scanning distance is linear with the QR code size as can be seen in the picture below. From this it can be concluded that the QR code should be as large as possible to be able to scan from as high as possible. Of course the maximum size of the code is limited because of various reasons. The landing pad can for example not block the whole street when waiting for the delivery. On top of that the landing pad should be easy to carry. Therefore the maximum size of the landing pad, and with that the code, is determined to be 1 x 1 meter. This because the drone itself will be at least 0.4 x 0.4 m due to the maximum size of the packages and a small clearance space is needed in case the drone will not land perfectly in the center of the landing pad.
Secondly it was noted that the influence of the used scope to zoom in was not always the same. The manufacturer of the scope said it should be up to eight times, but in reality it turns out to be much less. This is because the scope does not automatically focus. Therefore when using a scope on the drone to increase the maximum height at which the code can be scanned some changes have to be made. First the drone should be able to recognize a white square (on which it cannot yet scan the code) as a possible landing pad. Secondly the scope used to zoom in should have an auto-focus function as to guarantee that the code to be scanned can always be read clearly. If this turns out not to be a landing pad the drone looks for another white square. If the white square is a landing pad starts do execute its landing routine. The landing procedure described before thus changes a bit. Instead of flying around and directly scanning the codes on the pads the drone would first scan for a white square which is a possible landing pad and then focus on it to scan the code.
Thirdly the influence of the pixel density was as expected. The scanning distances has a negative quadratic polynomial relation with the pixel density as can be seen in the picture below. When the density increases the maximum scanning distance decreases. The fact that the relation is quadratic instead of linear is because the scanning distance is plotted against the width of the code in pixels. The total amount of pixels, which is what influences the maximum scanning distance, is the square of the width. From this follows that the lowest pixel density is preferred as described before.
Lastly the maximum angle at which the code can be seen by the drone, and thus scanned, was determined. From this it was found that the average angle lies between 27 and 30 degrees from a perpendicular line to the code. This means that the total field of vision is 54-60 degrees. When using the scope this field of vision narrowed to 10-16 degrees, severely limiting the area that can be scanned. This should be taken into account when designing the camera module on the drone.
All the experiments were done under ideal circumstances. When the circumstances were changed, such as bad lightning or a damaged code, the maximum scanning distances decreased. This should also be taken into account when determining the height at which the drone looks for the code, as in reality the circumstances will rarely be optimal.
From the experiments the following scaling factors for the phone with the best results, the Oneplus 2, can be derived:
[math]\displaystyle{ C1 = \frac{27.4w + 0.435}{r} }[/math]
In which w is the width of the code in meters and r is the reference distance
[math]\displaystyle{ C2 = \frac{0.0062m^2 - 0.5332m + 14.098}{r} }[/math]
In which m is the amount of pixels in the code in one direction.
The maximum angle at which the code could be scanned was 28 degrees. Therefore the maximum height at which the drone should fly is:
[math]\displaystyle{ Height = C1 \cdot C2 \cdot cos(28) \cdot R }[/math]
In which R is the real size of the landing pad.
The reference distance is chosen as 5.92 meter, the reference width of the code as 0.2 meter and the amount of pixels as 21. These values are chosen because they are present in every measurement conducted. Scaling this up to the landing pad of 1 meter diameter with the lowest amount of pixels, 15, as described before the maximum height at which the drone can scan the code is 32.9 meter. Changing the cosine in a sine gives a formula for the radius of the circle in which the pad can be detected, which is 16.5 meter for the chosen parameters.
Conclusion
From the experiments it could be seen that the unmodified cameras from the tested smartphones are not capable to scan a QR code from more than 36 meters. Therefore to increase the scanning distance a scope will have to be used. When using the scope during the experiment a larger distance could be reached, but at the expense of the maximum scanning angle. This was mainly because the scope covered a part of the camera, which could also be seen in the camera app. A better scope will thus have to be used in the drone. A perfect scope would allow scanning from greater heights without reducing angle or the detection radius. No such scope does exist of course, but in the detection range found during the experiment there was a margin of 6.5 meters compared to the minimum scanning radius of 10 meters to cover the GPS error. This means it would be possible with the right scope to scan from a high enough height and still maintain the radius.
Technical advice
This paragraph will explain some technical aspects about landing a helicopter and taking off again. It is a design proposal which contains information about how a helicopter delivery drone can be built, without specifying certain parts as it depends on different factors. During a landing session or a flight session, it is important that the drone is stable at all times. Also, for a delivery drone a switchable battery is desired, as this removes downtime of the drone caused by charging its battery. Also, the drone needs to be equipped with several sensors in order to be able to fly autonomously.
Control
Autonomous drones should always stay connected with the distribution center as it needs to give signals whether the delivery is going well or whether a task can not be completed. However, wireless connections sometimes experience bad stability, which causes connection losses. During those losses the drone must be able to get into a safe modus, a modus where it stops moving and keeps stabilizing itself. It should also be able to fly back to the distribution center if a new connection can not be established. Also, the drone may experience problems while trying to perform tasks autonomously. For that reason, it should always be possible for a human controller to connect with the drone. This requires multiple control systems and the design of these systems can be done in multiple ways.
First of all the drone should be able to stabilize itself and fly back to the distribution center when a connection is broken. The hardware for this control system should be implemented in the flight controller, which is located on the drone. The system should be intelligent enough to fly to a certain GPS location.
Secondly, it should maintain connection with the distribution center to send feedback on how it is performing its tasks. Also, software should be implemented to allow video streaming and full control of the vehicle by a human controller located in the distribution center.
Center of gravity
For optimal stability the center of gravity has to be as low as possible. A center of gravity which is lower than the propellers causes gravitational forces to automatically compensate for small disturbances. The center of gravity should never be above the propellers as this increases the disturbance which means the system will be naturally unstable. Unstable systems require more power because they need active stabilization.
Also the weight distribution is an important parameter, a wide distribution of weights increases the angular momentum. Increasing the angular momentum will make the helicopter less agile and will also make the helicopter less vulnerable to disturbances. However, greater angular momentum will increase power consumption while turning, as more energy is required to change its rotation.
Sensors
Autonomous helicopters require multiple sensors in order to percept their environment as they cannot receive all the required information from the main server. The required sensors for such a drone have been listed below.
- Flight sensor requirements
One of the most import sensors for a helicopter is a gyroscope. Gyroscopes provide information about angular velocity of the drone, which is required to compensate for disturbances on the drone which allows a stable flight. Also accelerometers can be added which allows measuring the linear velocity in different directions. This can be used to calculate flight speed, or compensate for unwanted disturbances. Also a barometer should be required, as this allows the drone to measure its altitude. A small on board computer with a stable radio or internet connection is also required. A 4G connection is easier to implement as the infrastructure has already been built by Dutch telecom companies. However, as connection protocols for radio connections are usually faster, a radio connection could also be used. Both ways can be used cover the entire country. One disadvantage of a radio connection is that the bandwidth is usually low. This does generally not support video streaming which is a requirement.
- Autonomous system sensor requirements
In order to navigate autonomously, a global positioning sensor will be required. This way the drone can locate itself and it can follow a flight path when combined with gyroscopes and accelerometers.
- Autonomous landing sensor requirements
For autonomous landing several sensors will be required. One important sensor will be a camera because the drone will need to be able to detect its environment to make sure that it does not land on an undesired location. Also proximity sensors can be used to detect whether it is close to colliding with an object or with the ground. Additional sensors may be required depending on the used method for landing and the design of the drone. An infrared camera can also be very useful as this allows the robot to detect persons or animals easily. In this situation the drone can react by hovering above the landing spot, while waiting till the person or animal moved away to a safe distance. In a situation where the landing spot remains obstructed the drone will be forced to fly back to its distribution center.
Battery
In order to pick the right battery size, the distance to the customer has to be determined. As the spread of distribution centers and the maximum flight range of the drone is not the focus on this project, this will be left blank. However, a suggestion can be made on the way of how to decrease downtimes. Be using a switchable battery system the drone does not have to stay at the distribution center to recharge but instead it can swap batteries and deliver the next package. Also, lithium-ion batteries are advised, as they can provide the highest power amongst batteries and they also have the best energy density. Picking the distribution center spread and therefore the desired battery capacity needs some further research.
Package safe
As described by PRE group 1, a package safe will be required to prevent theft of the package. NFC will be used to open the safe on the drone, which only allows authorized persons to open the safe. This makes stealing a package much harder.
Package location
One of the heaviest components in a drone is its battery. For this project packages of two kilograms will also be carried by the drone, making the battery and the package the most heavy components. After landing, the customer will need to be able to pick up the package in an easy way. Assuming the drone will land and turn off its propellers, it will be logical to place the package on the top of the drone. This makes it easily accessible. However, placing it on top of the drone raises the center of gravity, which might give stability issues. This can be solved by placing the battery at the very bottom of the drone, or by raising the position of the propellers.
Landing mechanism
For the package delivery drone it is important to land so that the package can be delivered in the most secure way. In order to land stably, without falling on the side and potentially damaging the package and the drone it needs proper landing gear. For the landing of a the drone it is most practical that it can land on a small surface and directly from above the landing destination, in this case a landing pad. If it can land from directly above the drone does not need a lot of landing space and can avoid most of the potential obstacles in the vicinity. Because propeller drones can fly or hover at a very slow speed they can also afford to land in this way and drones can speed up and slowdown in the air quite easily as well. Landing legs seem to be the better option for these kinds of drones and not coincidental most of the propeller already have, if anything, legs for landing gear. There are several advantages that landing legs can provide:
• The legs can be placed closer to the middle of the drone so that the landing pad does not necessarily have to be bigger than the drone itself. As long as all the legs fit on the pad the drone can land on the pad. (In this project the landing pad is still bigger than the entire drone for safety reasons).
• The legs have to be strong enough to support the weight of the drone, but they do not have to be very big in order to do that. Smaller or thinner legs do not have a lot of friction with the air. Also small legs may be retractable so that the friction is reduced further while the drone is flying normally.
• The landing legs could incorporate a mechanism that softens the landing of the drone for more safety. But this function may be trivial because a propeller drone is likely to already land very stable.
Currently Amazon is investigating landing legs that can even have different lengths to land stably on bumpy terrain [21]. This investigation also incorporate the ideas of absorbing the landing shock as well as additional mechanisms that secure a stable footing for the drone including building spikes, suction cups or even magnets.
In this project however these adjusting landing legs are not considered, due to the fact that the drone lands on the flat surface for a landing pad. The landing pad is flat anyway in order to read the QR code on the pad. There can also be different options besides legs. An example is an air cushion or normal cushions. They can provide a very soft landing but may be quite large and cause much friction in the air. There are not many examples of other concepts like this, so this project will not include more speculation about the landing gear and assumes that the drone has legs and those legs can land on the landing pad.
In the drone legs there are also several different concepts. The following figures illustrate some common examples.
General conclusion and discussion
In this chapter a general conclusion and discussion can be found. The chapter has been devided into the tecnical advice, control, center of gravity, sensors, battery, package safe and package location. The goal is to give a clear view on the implication of the conducted experiments into the future autonomous package delivery system.
References
[1] The self driving delivery robot, visited on the 8th of March 2017, visited from: https://www.washingtonpost.com/news/innovations/wp/2015/11/02/this-self-driving-delivery-robot-is-coming-soon-to-sidewalks/?utm_term=.3677b2f3536e ;
[2]The current prices for postal delivery, visited on 8th of March 2017, visited from: https://www.postnl.nl/tarieven/tarieven-pakketten/ ;
[3]The airship 600, visited on 16th of March 2017, visited from: https://en.wikipedia.org/wiki/Airship_Industries_Skyship_600 ;
[4]Costs of the Airship 600, visited on 16th of March 2017, visited from: http://www.newsday.co.tt/news/0,153096.html ;
[5] Autoriteit Consument & Markt, Pauline Gras, 20-05-2016, Nieuwsbericht: Online shoppen zorgt voor explosieve groei pakketvervoer Link: https://www.acm.nl/nl/publicaties/publicatie/15814/Online-shoppen-zorgt-voor-explosieve-groei-pakketvervoer/
[8] Volkskrant, niemand, 18 april 2016, nieuwsbericht: Meer mensen in stad dan op platteland http://www.volkskrant.nl/binnenland/meer-mensen-in-stad-dan-op-platteland~a767483/
… https://www.landenweb.nl/nederland/bevolking/
[9] Link: https://cdn.techinasia.com/wp-content/uploads/2016/10/BN-LM642_amzndr_G_20151130162550.jpg
[11] ... http://blog.qrstuff.com/2011/11/23/qr-code-minimum-size
[12] FAA anounces commercial drone rules, visited on 8th of March 2017, visited from http://spectrum.ieee.org/automaton/robotics/drones/faa-announces-commercial-drone-rules
[13] Final rules of the FAA for unmanned drones, visited on 8th of March 2017 ,visited from https://www.faa.gov/news/press_releases/news_story.cfm?newsId=20515
[14] Roadmap for drone operations in the European Union (EU), visited on 8th of March 2017, visited from https://www.dronewatch.nl/wp-content/uploads/2016/06/EASA_EU-Roadmap-Operation-Centric-Approach-Drone-Ops-v13_160620.pdf
[15] Explanation Civil Drones defined by EASA, visited on 8th of March 2017, visited from https://www.easa.europa.eu/easa-and-you/civil-drones-rpas
[16] ‘Prototype’ Commission Regulation on Unmanned Aircraft Operations defined by the EASA, visited on 8th of March 2017, visited from https://www.easa.europa.eu/system/files/dfu/UAS%20Prototype%20Regulation%20final.pdf
[17] Concept of Operations for Drones defined by the EASA, visited on 8th of March 2017, visited from https://www.easa.europa.eu/system/files/dfu/204696_EASA_concept_drone_brochure_web.pdf
[19] PostNL overview of 2016 - 2017, visited on 15th of March 2017, visited from ? is een pdf die ik ergens heb gevonden en gedownload.
[20] Drone flying school, visited on April 3th 2017, visited from: https://www.drones.nl/nieuws/2015/11/nlr-eerste-erkende-vliegopleiding-voor-drones-rpas
[21] Amazon landing legs for drones, visited on April 10th 2017, visited from: http://www.geekwire.com/2017/amazon-patents-drones-landing-legs-propellers/