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

Dataset Recordings

RLRoverLab records datasets from examples/03_inference/eval.py when --dataset_name is set. Files are written to <dataset_dir>/<dataset_name>.hdf5 on environment close.

python examples/03_inference/eval.py \
  --task AAURoverEnvRGBDRawWVGA-v0 \
  --num_envs 1 \
  --steps 1000 \
  --enable_cameras \
  --dataset_dir ./datasets \
  --dataset_name rover_wvga_expert_1000 \
  --dataset_type RL_COMPRESSED

Recorder Types

--dataset_typeFormatUse when
RLLegacy Isaac Lab HDF5, robomimic-style layoutA loader expects /data/demo_*/obs, /data/demo_*/next_obs, actions, rewards, and dones.
ILLegacy Isaac Lab HDF5, robomimic-style layoutOnly observations and actions are needed.
RL_COMPRESSEDOptimized RLRoverLab RGB-D schemaRGB-D storage size and random-access offline loading matter more than direct robomimic layout compatibility.

Legacy HDF5

RL and IL use Isaac Lab’s default HDF5 dataset writer. The layout follows the robomimic convention of storing demonstrations under /data/demo_N. Current robomimic training compatibility still depends on the loader, environment metadata, and observation keys used by the training config.

For RL, each episode contains:

PathContents
/dataRoot data group. Attribute total is the total transition count; env_args stores environment metadata as JSON.
/data/demo_NOne recorded episode. Attributes include num_samples and optional seed and success.
/data/demo_N/actionsAction tensor for each transition.
/data/demo_N/rewardsReward tensor for each transition.
/data/demo_N/donesDone flags for each transition.
/data/demo_N/obs/...Observation at time t.
/data/demo_N/next_obs/...Observation after the action, at time t + 1.

IL uses the same episode layout but records only actions and obs. Datasets are gzip-compressed by HDF5. This format is simple and compatible with many robomimic-style loaders, but RGB-D trajectories are large because next_obs physically duplicates the next observation tree.

Use it with:

python examples/03_inference/eval.py \
  --task AAURoverEnvRGBDRawWVGA-v0 \
  --num_envs 1 \
  --steps 1000 \
  --enable_cameras \
  --dataset_dir ./datasets \
  --dataset_name rover_wvga_legacy_1000 \
  --dataset_type RL

Optimized RGB-D HDF5

RL_COMPRESSED writes RLRoverLab’s optimized RGB-D HDF5 schema. It stores RGB and depth observations once in an indexed timeline instead of duplicating a physical next_obs tree.

Use it when RGB-D storage size and random-access offline loading matter more than direct robomimic layout compatibility. The full dataloader contract is in Optimized RGB-D HDF5.

Record an optimized RGB-D dataset with:

python examples/03_inference/eval.py \
  --task AAURoverEnvRGBDRawWVGA-v0 \
  --num_envs 1 \
  --steps 1000 \
  --enable_cameras \
  --dataset_dir ./datasets \
  --dataset_name rover_wvga_compressed_1000 \
  --dataset_type RL_COMPRESSED