HF RL Explorer

Allow custom rollout buffer class for on‑policy algorithms

Allow custom rollout buffer class for on‑policy algorithms: a task in LegoFlow-SWE (Harbor dataset). Currently, the on‑policy algorithms A2C and PPO always pick the rollout buffer class automatically: - DictRolloutBuffer if the observation space is Dict , - RolloutBuffer otherwise.

The task

Currently, the on‑policy algorithms A2C and PPO always pick the rollout buffer class automatically: - `DictRolloutBuffer` if the observation space is `Dict`, - `RolloutBuffer` otherwise.

Part of Lego-X/LegoFlow-SWE.