`srb agent learn` — Learn from Demonstrations

Train offline from a dataset produced by srb agent collect. It runs without Isaac Sim.

Usage

srb agent learn --algo ALGO --env ENV_ID [options]

Options

Option	Description	Default
`-e`/`--env`/`--task` ENV	Environment ID for log directory and specs	REQUIRED
`--cfg` PATH	Hydra YAML override; `DEFAULT` uses environment defaults, `IGNORE`/`NONE`/`NULL` disable it	`DEFAULT`
`--algo` ALGO	Imitation Learning algorithm	REQUIRED
`--dataset` PATH	HDF5 dataset; if omitted, searches the new run directory and errors if none is found	`None`
`--num_epochs` N	Number of training epochs	algo default
`--normalize_actions`	Normalize actions in the dataset	`False`
`--logdir`/`--logs`	Root directory for learning run outputs	`SRB_LOGS_DIR`

Dataset admission and normalization

Training reads a verified private HDF5 copy. Saved configuration records the source path but does not archive the dataset.

Admission rejects inconsistent sample counts or terminal flags and filters that select no demos. Native configs may select custom action keys.

--normalize_actions preserves the source and records ranges in normalization_params.txt; constant-action and no-action inputs produce none.

Supported Algorithms

Offline learning uses robomimic; names start with robomimic_:

Algorithm	Description
`robomimic_bc`	Vanilla Behavioral Cloning (MLP policy)
`robomimic_bc_rnn`	Behavioral Cloning with RNN policy
`robomimic_bc_transformer`	Behavioral Cloning with Transformer policy
`robomimic_bcq`	Batch-Constrained Q-Learning
`robomimic_cql`	Conservative Q-Learning
`robomimic_iql`	Implicit Q-Learning
`robomimic_td3_bc`	TD3 with BC regularization
`robomimic_gl`	Goal-conditioned Learning
`robomimic_hbc`	Hierarchical Behavioral Cloning
`robomimic_iris`	IRIS (latent skill discovery)
`robomimic_diffusion_policy`	Diffusion Policy

See the Robomimic documentation for algorithm details and hyperparameters.

Examples

Train a Behavioral Cloning policy on a dataset collected for peg_in_hole:

srb agent learn --algo robomimic_bc --env peg_in_hole \
  --dataset logs/peg_in_hole/collect/<run>/datasets/demos_<timestamp>.hdf5

Train on a specific dataset path with an explicit epoch budget:

srb agent learn --algo robomimic_bc_rnn --env peg_in_hole \
  --dataset /path/to/demos.hdf5 --num_epochs 500

Output

Checkpoints and metrics go to <logdir>/<env_id>/<algo>/<timestamp>/; <logdir> is --logdir/--logs or SRB_LOGS_DIR. Evaluate the policy with:

srb agent eval --algo robomimic_bc --env peg_in_hole

or rolled out to produce additional demonstrations with:

srb agent collect --algo robomimic_bc --env peg_in_hole --num_episodes 200

Pass --model-artifact-output OUT to package the run as an SRB model artifact. The --dataset path supplies training-data provenance unless model-artifact dataset options override it.

Space Robotics Bench