All Tools
T
AgentsFreeOpen Source
TORCHRL
Modular PyTorch library for reinforcement learning primitives
MIT
ABOUT
Custom RL research often fights PyTorch data layouts, environment wrappers, and replay plumbing that do not compose. TorchRL gives TensorDict-based primitives for envs, collectors, and buffers so you can build single-agent, multi-agent, or offline pipelines that stay native to PyTorch.
INSTALL
pip install torchrlINTEGRATION GUIDE
1. Build a custom PPO or SAC trainer from TorchRL collectors, losses, and replay buffers
2. Run multi-agent RL with TorchRL's MARL environment and tensordict APIs
3. Train offline RL policies from datasets using TorchRL replay and loss modules
4. Integrate Gymnasium, DMControl, or custom sims into a PyTorch-native RL loop
TAGS
pythonpytorchreinforcement-learningtensorsreplay-buffermodular