The source code for our paper: Graph Reinforcement Learning Application to Co-operative Decision-Making in Mixed Autonomy Traffic: Framework, Survey, and Challenges. This code is developed based on our previous repository TorchGRL.
Hello, I noticed that in the learn_one_step() function of GRL_library/agent/continuous/DDPG_agent.py at line 258, the critic value is being detached. Typically, the DDPG algorithm uses detach() on the critic_target value, so this might be a typo. Also, it looks like the AC (actor-critic) method might be missing the detach() function for the critic target value too.
Hello, I noticed that in the learn_one_step() function of GRL_library/agent/continuous/DDPG_agent.py at line 258, the critic value is being detached. Typically, the DDPG algorithm uses detach() on the critic_target value, so this might be a typo. Also, it looks like the AC (actor-critic) method might be missing the detach() function for the critic target value too.