PRE2015 3 Groep2 week3

From Control Systems Technology Group
Jump to navigation Jump to search

Week 3: Redefining focus and deliverables and starting research

Redefined focus

The focus of this project is into the last few meters. To deliver the package to the front door the process can be divided in three steps. The finding, the communication and the flying away. With each step, there are some problems that need to be solved.

Previous research, done in the second quartile of the year 2014/2015 by group 1 [1], answered some of the questions asked. They investigated the way of navigating and verifying. Their founding's, and some background information, will be discussed briefly in the sections below.

First step: Finding the costumer

One of the problems found during the delivery proces done by drone is finding where the drone can land and for which person the package is the drone is caring.

Location of the drone

An important factor of delivering a package is that it gets to its place of destination. There are several systems to provide this information. The most used and best know is the Global Positioning System (GPS).

A GPS-receiver uses a least four different satellites, in a direct line of sight, to determine its location. There are two different kinds of GPS: civilian GPS (SPS) and military GPS (PPS)[2]. The online difference between these two types is the frequency it uses to send out the signal. The GPS-signal can be received on the whole earth and an internet connection isn’t needed. The accuracy of the location determined with GPS depends on some factors. The most important factor it the quality of the receiver.

Since GPS uses time to calculate the position of the receiver and the receiver isn’t connect all the time to an atomic clock, the less a clock tends to drift the higher the accuracy will be of the calculated position. Most of the time it’s safe to say that, on solely using GPS, the position determined has an accuracy of 15 meters.

A newer system, and used alongside the GPS-system, is the Global Differential GPS System (GDGPS)[3], Which is developed by the Jet Propulsion Laboratory. This system uses ground based sending station to send out the error of the position of the satellites of the GPS-system. This improves the position determined with GPS from 15 meter to 10 centimeters.

Finding the costumer

Determining the position of the costumer is done with the use of GPS. The user has an app installed which communicated with the servers of the company sending the package. This app will update the whereabouts of the costumer to the server. The drone can then retrieve these GPS-coordinates to determine where the costumer is. When the drone is hovering at the right spot, found with the given GPS-coordinates, the drone will decent vertically until it is save on the ground. The drone will land with an accuracy of 3 meters on the right spot.

Delivering to a front door which can't be reached

It may be possible that the drone can't reach the front door of the customer. This can happen when, for example, the costumer lives in an apartment complex where the front is covered or the front door is at the inside of the building. Since the drone needs to land and can't enter the residence of the costumer, the delivery can't be done through the window.

At the moment deliveries in apartment buildings, done with real humans, have different solutions for this problem. Some companies expect the costumer coming down to the front door of the building, other companies let the mailman go to the front door of the apartment. The first may be a solution for the drone delivery problem, the later isn't.

Second step: Releasing the package

The drone is equipped with an NFC-reader. This reader will read the NFC-tag send out by the mobile phone of the costumer. This tag is generated by the server and send to the app mentioned earlier. If the tag send out by the user, and received with the NFC-reader mounted on the drone, is the same as the tag generated by the server, the package will be released.

Focus during the project

Two important and interesting problem remain. How do we get the flying more autonomous, so the accuracy of the landing spot is better, and how do we get the interaction with the people as human friendly as possible.

The first question asked will look into the problem of finding the right landing spot. Typical questions asked during this part of the project are: where is the costumer actually and are their any inanimate objects on the planed trajectory path.

Question asked during the part where the user will be central, the interaction on a human friendly way, will be for example: what is a comfortable landing distance for the user and what is the most comfortable landing path for the costumer.

Redefined requirements

  • The drone can decided what a good landing spot is
    • Their is enough free space
    • The drone lands on a comfortable distance of the user
  • The drone is able to land autonomously
    • The drone is able to evade inanimate objects
    • The costumer or drone won't be endangered when something doesn't go as planned, for example when the landing spot is uneven
  • The drone follows a trajectory path which is comfortable for the user


Research has to be done to get the right procedure for the way of landing with a drone. How to autonomously land with a drone? Finding the right spot to land and finding the right way of approaching the customer with a drone.


Landing the drone is thought to be easy, simply ascent till you hit the ground and that's it. This however, is not that easy[4][5][6]. Questions arise like; how does the drone avoid objects on the ground? And how does he find the right spot to land? Is having the customer find an open space and put a printed "A" on the ground (like Amazon's Prime Air[7]) a good solution, which makes the customer responsible. Or should the drone find his own way to find the right spot?

Giving customers a lot of responsibility for the landing of the drone, is not a good idea. The company that uses the drone is responsible for the package until the customer verified that the package is in their hands. This way it gives a great risk for the company for using a printed label like Amazon wants to do. Since this is not an option, the way of finding a spot to land can be a good option. Measuring the height of the drone according to the surface below and storing this, can be converted to a height map. If there is enough space on the map on a lower area where the drone can fit, it can land there. For this a safety margin can be put for an extra 10 centimeters on the sides of the drone.

But what if, for example, the drone chooses to land in the middle of a road? Considering this is probably the largest free space available close to your front door, this will be a very likely option. How does the drone distinct places where he is not able to land?

Another thing that has to be kept in mind are the (unexpected) moving objects. This includes among others footballs, birds and children. Seeing objects with a drone is really hard. The system has to identify how far something is away, which is for example impossible if you can record in only 2 dimensions with a camera. However, a lot of research about object avoidance is done these days (for example: [8] and [9]) and besides the landing part it is not the main scope of this project.

Approaching the user

So far little research has been done on the field of approaching humans, when it comes to robots. Even less so for drones in specific. The behavior-based navigation architecture is one way of how robots can decide which way to approach people. Previously done research by E. Torta[10], regarding approaching people with robots, gives a good insight and starting point. Based on the results of these experiments a model of a person's personal space concerning the Nao robot was made. After that a smart algorithm was made to find the optimal spot for communicating, while keeping in mind obstacles that could block certain positions and or routes.

On one hand drones give a extra dimension to this research, since also height should be implemented. On the other hand the robot used in the experiments described by Torta first approaches people and then seeks, for the user will be attending other busyness at that moment. The delivery drone we are talking about however will have the attention of the user from the start, meaning the aspect of orientation of the user can be left out.

Combining this approach with the previous subsection "Landing", the right landing procedure can be made. First the approach of the drone is used and when the drone is on the right spot to land it is going to look if it can land. If it can't land it should go to the next best place and so on and so on. To get this procedure validated for actual use, the Landing and approach is put in a test situation with a drone. This is explained in the chapter Technical implementation.


  1. J. Boonen, L. de Jong, R. Kerstens, J. Kruijtzer, and J. Linssen. 'Flexible drone delivery', 2014.
  2. “GPS Accuracy”
  3. Jet Propulsion Laboratory. “The Global Differential GPS System”
  4. 'SpaceX Launch Successful, But Drone Ship Landing Fails'
  5. TU Delft, 'New theory allows drones to see distances with one eye'
  6. Youtube, AR-Drone autonomous takeoff and landing'
  7. Amazon introducing drone delivery. 'Amazon Prime Air', 2016.
  8. S. Crowe, 'MIT Drone Autonomously Avoids Obstacles at 30 MPH', 2015.
  10. Approaching Independent Living with Robots, February 2014E. Torta'