​

4 min presentation in DD2438

Group 16

Zhuocheng Wei and Sean O Riordan

​

https://docs.google.com/presentation/d/1b9sUVbee9HcOCUdIU25JQnDlNpUZXZqzc-GOeHSxiYE/edit#slide=id.p

Our progress so far on P1&P2

• Old: Using velocity obstacles for collision avoidance
• Define a set of valid velocities outside of all collision cones, assign a cost to each
• New: Using HRVOs
• Using motion model for the velocity obstacles
• Treating blocks as VOs with zero-velocity
• Using bounds to determine closest distance on the mesh
• Weighting costs by the distance to next point in path
• Using raycasts to dynamically update the path, incase we move backwards in the path

​

Car

Drone

Problems

• Agents get stuck looping back and forth, as they try to move forward in the path but have to move back to stay away from other agents (from the cost function)
• This could perhaps be fixed by using a priority system for each car, allowing one car to move forward if another stays still
• Agents get stuck on blocks, this problem not seen with other agents however
• Pathfinding system is unreliable, required a lot of alterations but still sometimes cannot find path

P2 - final maps - open, P1 failed similarly

P2 - final maps - semi open, P1 failed similarly

P2 - final maps - intersection, P1 failed similarly

P2 - final maps - highway, P1 failed similarly

P2 - final maps - onramp, P1 failed similarly

Our progress on P3&P4

​

​

• Statically setting one car/drone to the leader
• Old: Calculate how the cars/drones are positioned relative to the leaders velocity by projecting on it. (not anymore)
• Won't be able to tell whether the car/drone is ahead or not if the leader turns (see top right), the outer car will turn and treat the innerst car as being ahead even if when that car has not passed the gate.
• New: Calculate distance to the gate for each car/drone and adjust on that
• Taking the euclidean distance between gates, works pretty well on 6 and 6B since the A*path are straight lines.
• Does not well so well for cases with twisting road (top right), but the strict measurement between distances can make up for it quite a lot.
• Drones are a lot easier to be kept in formation, got way less penalties for them
• Capped target velocity below maxspeed gave longer completion time but much less of penalty

P3 6 car

P4 6 drone

P3 6B car

Ideally could’ve have some collision avoidance by speed up the cars in front a bit to let the last car pass

P4 6B drone

The big deviation for the outer drone can be prevented by just removing a condition check (having it right now only for a better runtime)

Progress Status Week 8

• Comment to customer paying 250 000kr for the report:
• “Being a short week, but still managed to work by fairly amount”
• Planned Time spent: 80%
• (Out of the combined 200h)
• Actual Time spent: 85%
• Out of the combined 200h
• Actual Progress: 75%
• (estimate progress towards completing assignment)
• Risk of not completing assignment: 8%

​