HF RL Explorer

Multiple Issues: Collector Info Loss, Gym Compatibility, and Dict Observation Normalization

Multiple Issues: Collector Info Loss, Gym Compatibility, and Dict Observation Normalization: a task in LegoFlow-SWE (Harbor dataset). Reproduction : Use a vectorized environment that returns info dicts on reset (e.g., with “final observation” or custom data). Collect episodes and inspect the…

The task

*Reproduction*: Use a vectorized environment that returns info dicts on reset (e.g., with “final_observation” or custom data). Collect episodes and inspect the `Batch.info` – it will be stale or empty.

Part of Lego-X/LegoFlow-SWE.