Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Integration — Imitation Learning

Train an offline imitation policy from teleoperation or policy-rollout demonstrations with srb agent learn --dataset <path> --algo robomimic_*. IL checkpoints cannot resume RL training. Validate hardware deployment separately.

Available Integrations

SRB integrates the following IL stack:

  • robomimic [PyTorch] - Reference implementation of BC (MLP/RNN/Transformer), BCQ, CQL, IQL, TD3-BC, GL, HBC, IRIS, and Diffusion Policy with a standard HDF5 dataset format.

Collected HDF5 datasets are consumed by srb agent learn --dataset with a robomimic_* algorithm. This page documents no Robomimic HDF5 replay-buffer loader for the integrated RL frameworks. Separate JSONL prefill paths exist for SKRL and Dreamer; see the training reference.

Data Format

Demonstrations are stored in the robomimic-compatible HDF5 layout:

  • Per-episode groups containing obs/, next_obs/, actions, rewards, terminated, truncated, and dones datasets (plus a per-episode success attribute).
  • Observation keys use nested state/<leaf> paths for SRB observation groups such as state, proprio, command, and privileged state.
  • Success flags are stored per-episode to support --success_only filtering at collection time.

See the collection output contract for the dataset path and learner handoff.

How to Get Started?

See Installation, Basic Usage, and the Imitation Learning Workflow.

See Also