sanshibayuan / multiagent_rl

Multi-Agent Reinforcement Learning with Particle Env. (on going)

Multi-Agent Reinforcement algorithms with Particle Environment (OpenAI) using pytorch

1. ddpg.py: bidirectional LSTM actor + LSTM critic + DDPG

2. model_ddpg.py: bidirectional LSTM actor + LSTM critic + DDPG + estimate next_state + estimate reward <estimate next_state + estimate reward> is a combined method between model-free RL and model-based RL.
   It shows significantly improved performance in particle env. simple_spread scenarios

3. model_rdpg.py: bidirectional LSTM actor + LSTM critic + RDPG (recurrent DPG) + estimate next_state + estimate reward This algorithm currently shows degraded performance than others in particle env. simple_spread scenarios.