Toolkit for learning driving models through maximum entropy inverse reinforcement learning, and autonomous vehicle's control through leverageing effects on human actions.

(Companion code to a paper presented at RSS 2016)

Running

To visualize: ./vis {file_name}.pickle

To run an experiment ./run {world_name} where world_name can be any one of the worlds defined in world.py

To run an experiment with irl_ground world: ./run irl_ground

To run the IRL algorithm: ./irl.py data/*.pickle

Modules

• dynamics.py: This contains code for car dynamics.
• car.py: Relevant code for different car models (human-driven, autonomous, etc.)
• feature.py: Definition of features.
• lane.py: Definition of driving lanes.
• trajectory.py: Definition of trajectories.
• world.py: This code contains different scenarios (each consisting of lanes/cars/etc.).
• visualize.py: This contains the code for visualization (GUI).