Ervin T.

[DRAFT][DON'T MERGE] Enable learning from multiple priors

1

### Proposed change(s) This PR adds a feature that was inspired by the approach used by DeepMind in [this paper](https://arxiv.org/abs/2105.12196). By specifying (i.e. giving the directory to that run id)...

[DON'T MERGE] Normalize GAIL encoder inputs

1

### Proposed change(s) Normalize the GAIL encoder inputs using incoming trajectories; this was done in TF but not in Torch. This helps stabilize training and improves performance on Crawler with...

Hot Load and Chunked Save of Demo Files

1

### Proposed change(s) Adds two new IL improvements: allow for adding additional .demo files during a training run, and saves new demos from the DemonstrationWriter in chunks of `Num Steps...

[bug-fix] Fix sequence length buffer truncate

1

### Proposed change(s) Fixed an issue where if 80% of the buffer isn't a multiple of sequence length, we'll get weird sequences in the buffer. Only affects SAC with LSTM....