ChenglongChen/pytorch-DRL

Provides modular, agent-based implementations of A2C, ACKTR, DQN, DDPG, and PPO with a unified interface supporting both single-step and n-step environment interactions. Each algorithm encapsulates core components—environment interaction, batch training, exploration/exploitation action selection, and evaluation—enabling code reuse across different DRL methods. Integrates with OpenAI Gym environments and uses PyTorch as the computational backend, with KFAC optimizer support for second-order optimization.

611 stars. No commits in the last 6 months.