multi_agent_path_planning
multi_agent_path_planning copied to clipboard
SIPP does not behave as intendet
Hello,
I tested the repository and did a number of experiments with the SIPP and CBS algorithms. I noticed that in many situations the SIPP algorithm produces a collision even if there would be enough space to avoid said collsion. The agents would also wait one time step before the collision as if the agent knows about the collision but then decides to no avoid it. I'll include a few examples. I explicitly tried to replicate the environment from the SIPP paper you referenced to show the difference.
How SIPP behaves:
Expected behaviour: