ucla-mobility / OpenCDA

A generalized framework for prototyping full-stack cooperative driving automation applications under CARLA+SUMO.
Other
1.02k stars 199 forks source link

Is reinforcement learning supported? #105

Closed xiaoyucheri closed 3 years ago

xiaoyucheri commented 3 years ago

Thanks for creating this great project! I saw OpenCDA has some rule-based algorithms as planning module. So does it also support reinforcement learning behavior planning? Thanks!

DerrickXuNu commented 3 years ago

Yes, OpenCDA supports RL development. Pass the RL states to the behavior agent through function update_information(), execute the action in run_step() and collect the rewards in plan_debuger() or EvaluationManager(). If you are using deep RL, you can put your model in MLManager and add it as an attribute in your behavior agent class. Check https://opencda-documentation.readthedocs.io/en/latest/md_files/customization.html to see more details.

xiaoyucheri commented 3 years ago

That works, thanks!