PRE2019 4 Group8

Luc Geurts, Tar van Kieken, Sietse Backx, Mandy Grooters, Rien Boonstoppel

Solution Concepts

Our solution can be split into 2 seperate components:

• Tracking what parking places are available
• Guiding user vehicles to parking places

For both of these categories we present and discuss several possible solutions.

Tracking parking spaces

Offline scan

This solution would perform a scan of the parking lot at the start of the day and check what spots are available. Afterwards, it will keep track of available spaces virtually. A spot will only be taken if it was already taken at the start of the day, or the system itself assigned it to a vehicle during the day. This technique would uses the assumption that when a spot is taken somewhere during the day, it will remain occupied throughout the day. If this is not the case, the system will waste parking spots. In addition, the system could perform another scan during the day to catch the spots that might have become available, when the system thinks all spots are occupied.

The scan of the parking lot itself can be rather simple in this case, since it doesn't have to be very efficient (as it doesn't happen often). Two feasible options are:

• Manual 'scanning' by employees. This could be rather feasable in some specific situations, such at our running example of the Efteling;
  • At the start of the day (before opening) there are barely any cars on the lot
  • When employees have to rescan when the system thinks no spaces are left, there will probably not be many empty spaces
• Scanning by autonomous ground vehicles. A ground vehicle, which might be used to guide cars as well, may be used to go over each spot in the parking lot and check if it's taken. If it passes each spot, it could simply use a distance sensor to decide whether a car is present. It may also use a camera with simple computer vision to achieve the same, but this might already be slightly more complex.

Online per spot tracking

A sensor could be used for each parking spot individually to determine whether a car is present. Multiple sensors could be used to achieve this:

• A weight sensor within the parking space to detect the weight of a car
• A distance sensor at the end of the parking space, to detect whether something is within a certain distance

This way the system could easily and accurately see what spots are available at any given time, without having to make any assumptions.

Online global scan

Cameras with computer vision could be deployed to track whole areas of the parking lot at once. This would involve installing enough cameras to cover the whole parking lot, and making use of computer vision to detect whether a spot is taken at any given time.

Semi-Online scan

An autonomous aerial vehicle could regularly fly over the parking lot and scan what spaces are available, using a similar camera setup as with global tracking. The aerial vehicle could cover the area quite quickly compared to ground vehicles and thus makes it possible to perform these scans many times an hour. Moreover, since these are aerial vehicles, the scans don't interfere with car guiding operations at all.

Guiding vehicles

Following ground drones

An autonomous ground vehicle could be deployed to physically guide a vehicle to their designated parking spot. This vehicle would start in front of the car that needs to park, and drive towards the assigned parking spot, somehow signalling the user vehicles to follow. This would requite the autonomous vehicle to be large enough to be noticed by the user vehicle, and not be run over. In addition it should drive fast enough to have a pleasant speed to be followed by user vehicles. Since this process would be rather slow, multiple of these robots should be deployed to guide vehicles in parallel. These vehicles need to somehow return back to the queue after having assigned a user vehicle to their spot. This would require a route to the start that doesn't interfere with guidance of other vehicles.

Following air drones

An autonomous aerial vehicle could be deployed to physically guide a vehicle to their designated parking spot. This would work similar to the ground drones, but have one additional problem. Since these are aerial vehicles, it's probably difficult to make them large and remain safe, thus it becomes more of a challenge to make them stand out and noticable by the user vehicle. For these vehicles, lights can be used to grab user attention instead. Having return paths becomes easier for these vehicles, compared to ground vehicles, because they could simply be flying at an other attitude to prevent collisions.

Following ground drone instructions

In certain cases, when the parking lot is rather simple/structured, a pair of autonomous ground vehicles could be deployed to point indicate the parking spot to be used. If the parking spot is a simple structured grid, any parking spot can easily be indicated by showing the row and column of the spot. Ground vehicles could do this physically by stopping at the row and column to be parked in. One of these vehicles would always remain on the main path and indicate the row to be parked in. The other vehicle would be present within said row and indicate the column to be parked in. These autonomous vehicles should be provided with a means of pointing the user vehicle in a direction to move, since user vehicle is not expected to physically follow the vehicles in this situation. The autonomous vehicles would simply move to the next column and or row whe a spot is taken.

Solution Concepts Discussion

None of the suggested solutions is perfect, so below is a discussion of the pros and cons of each approach. For each of the solutions, it's also mentioned how feasible it would be to develop the technology by our group.

Tracking parking spaces

Offline scan

Pros:

• Can be very simple to deploy in case:
  • Employees are used and the parking lot is always almost completely full or empty at time of the scane.
  • Ground vehicles are also used for guiding vehicles already.
• Doesn't require any electronics to be permanently installed.

Cons:

• Relies on the assumption that vehicles primarily leave at the end of the day.
• Scanning may take a considerable amount of time, and may block the guidance system from functioning.

Feasibility:

• May not require any electronics when relying on employees (and is thus entirely feasible).
• May reuse behavior that must already be present in guidance by autonomous ground vehicles, and thus not present any new challenges.

Online per spot tracking

Pros:

• Gives an entirely accurate live overview of the parking lot.
• Is very robust, due to the simple sensor setup.

Cons:

• Requires installation of each of the parking spaces of the parking lot.

Feasibility:

• Requires only very simple sensor usage, and thus be very feasible.

Online global scan

Pros: - Gives a live overview of the parking lot. Cons:

• Requires installation of several cameras throughout the parking lot.
• Is not entirely reliable due to usage of computer vision, which is error prone.

Feasibility:

• Requires usage of computer vision, which can be challenging.

Semi-Online scan

Pros:

• Doesn't require any installation.
• Gives an almost live overview of the parking lot.

Cons:

• Is not entirely reliable due to usage of computer vision, which is error prone.

Feasibility:

• Requires usage of computer vision, which can be challenging.
• Requires autonomous operation of an aerial vehicle, which can be very challenging:
  • Usage of GPS for approximate location tracking, which is doable, but may not be accurate enough to target exactly 1 parking space.
  • Requires some very precise autonomous control in order to be docked, which is very difficult.

Guiding vehicles

Following ground drones

Pros:

• Can be a large vehicle, with large capacity batteries, thus not requiring an autonomous docking solution.
• May be easy to understand by user vehicles (research would need to be conducted).

Cons:

• Requires many vehicles to be efficient.
• Wastes a lot of resources:
  • Requires to drive all the way from the start of the queue to the designated parking spot per user vehicle, costing a lot of energy.
  • Requires to drive all the way back to the queue, during which it's not of use to anyone.
• Requires a separate return path that doesn't interfere with the other vehicles being guided.

Feasibility:

• Requires autonomous operation of a ground vehicle, which can be challenging but should be achievable:
  • Usage of GPS for approximate location tracking, which is doable.
  • Usage of computer vision for more accurate obstacle avoidance, which should be doable.
• Requires usage of computer vision, for checking whether user vehicles are parked and for obstacle avoidance, which can be challenging.
• Can be challenging to coordinate multiple autonomous vehicles at once.

Following air drones

Pros:

• Doesn't require a special return path, a different altitude can be used instead
• Doesn't have a lot of wasted time by returning to the queue, due to the speed fyling vehicles can easily have.

Cons:

• Requires many vehicles to be efficient.
• Wastes a lot of resources:
  • Requires frequent recharging due to the small battery capacity in order to be light enough to fly.

Feasibility:

• Doesn't require active obstacle avoidance, since there are only other drones in the air (which should just be properly coordinated)
• Requires autonomous operation of an aerial vehicle, which can be very challenging:
  • Usage of GPS for approximate location tracking, which is doable, but may not be accurate enough to target exactly 1 parking space.
  • Requires some very precise autonomous control in order to be docked, which is very difficult.
• Requires usage of computer vision, for checking whether user vehicles are parked, which can be challenging.
• Can be challenging to coordinate multiple autonomous vehicles at once.

In short, when compared to following ground drones, it's easier in the sense that no active obstacle avoidance is needed, but harder in the sense that it requires docking.

Following ground drone instructions

Pros:

• Only requires 2 ground drones.
• Doesn't require the drones to cover a large distance (and are thus relatively energy efficient).

Cons:

• Requires the parking lot to have a really specific and simple layout.
• Has to have a clear way of signalling instructions to cars
• Can not easily fill random spots that have become available
• Has some reset time once it hit the end of the parking lot, and may even pose difficulties if the entire queue already moved along to the last row

Feasibility:

• Requires autonomous operation of a ground vehicle, which can be challenging but should be achievable:
  • Usage of GPS for approximate location tracking, which is doable.
  • Usage of computer vision for more accurate obstacle avoidance, which should be doable.
• Requires usage of computer vision, for checking whether user vehicles are parked, which can be challenging.

In short, when compared to following ground drones, it's easier since it doesn't require complex coordination of multiple drones.