replay-dataset

Replay a recorded LeRobot dataset episode on the robot.

Replays one episode of an existing LeRobot-format dataset on Axol: the episode’s recorded action column is streamed back to the arms frame by frame, then the robot returns to its rest pose. It’s the inverse of collect-data — instead of recording teleop actions it plays an already-recorded episode straight onto the hardware.

No cameras, teleop, or policy server are involved: replay only drives the arms, so the robot config carries no cameras. The robot is first moved to the rest pose (the same collision-aware return-to-rest run-policy uses) so the arm starts where every recorded episode does — episodes are recorded from rest, so the first replayed action is ~rest and there’s no jump. Each frame is sent at the dataset’s recorded fps to reproduce the original timing — paced on the robot’s control loop so command intervals stay regular and the motion stays smooth — and a final return-to-rest leaves the arm parked. With --interpolate the recorded actions are upsampled to ~120 Hz commands for even smoother tracking, without changing the episode’s timing.

Pass --loop to replay the episode continuously (returning to rest between takes) until you stop it with Ctrl+C (or Stop in the control panel); either way the arm returns to the rest pose before the operation exits.

This command is configured via draccus. The robot subsystem is exposed as the nested robot_config config (CAN channels, per-joint gains) — nest into it with dots or pass a whole-config file. See Command configuration.

Flag	Description
--repo_id <id-or-path>	Dataset to replay (required): a repo id under --root / $HF_LEROBOT_HOME (as recorded by collect-data, e.g. myorg/pick-place), a HuggingFace Hub repo id to pull, or a filesystem path to a dataset directory (one containing meta/info.json) anywhere on disk. When the dataset isn’t found, the error lists the datasets present on the machine.
--episode INT	Index of the episode to replay, 0-indexed — the first recorded episode is 0 (required)
--fps INT	Playback frame rate. 0 (default) replays at the dataset’s recorded fps, reproducing the original timing; set a positive value to override it.
--interpolate true	Smooth playback: linearly interpolate between recorded actions and command the arms at ~120 Hz (the teleop control rate) instead of the dataset fps. Episode timing is unchanged — only the command granularity increases. Default: off (each recorded action is sent once, as-is).
--loop true	Replay the episode continuously (returning to rest between takes) until stopped with Ctrl+C / Stop, then return to rest. Default: off (a single replay).
--reset_torque_threshold FLOAT	Contact watchdog for every return-to-rest (default: 4.0 Nm). A joint torque residual (measured minus modeled gravity) sustained above this means the move hit something — it stops and the arms drop into a limp gravity-compensation hold. Replay has no interactive retry, so free the arms and stop the run (Ctrl+C / Stop). 0 disables the watchdog.
--reset_gravity_comp_kd FLOAT	Velocity damping (Nm·s/rad) for that contact-fallback hold (default: 0.25); same semantics as gravity-comp --kd.
--teleop_torque_threshold FLOAT	Contact watchdog for the playback phase — checked on every command as the episode plays, separate from the return-to-rest guard above. Off by default (0), since a replayed episode touches the scene on purpose; set a threshold (the control panel suggests 16.0) and a sustained torque residual past it stops playback and drops the arms into the limp gravity-comp hold until the run is stopped (replay has no interactive retry). Mirrors teleop’s field of the same name.
--root PATH	Local dataset root (default: $HF_LEROBOT_HOME)
--robot_config.axol_config.left_stiffness S	Compliance↔stiffness blend for the left arm in [0, 1]. Scalar or 7-element list (one per arm joint, in Joint enum order). 0 = fully compliant; 1 = pre-tuning industrial gains; 0.5 (default) is the geometric mean. Use right_stiffness for the right arm. Match the stiffness used at data-collection time so the arm tracks the recorded actions the same way. See AxolConfig.left_stiffness.
--log_level {DEBUG,INFO,WARNING,ERROR}	Default: INFO
--config_path PATH	Load a whole-config JSON/YAML file; CLI overrides layer on top.

axol replay-dataset --repo_id myorg/pick-place --episode 0
axol replay-dataset --repo_id myorg/pick-place --episode 0 --loop true   # loop until Ctrl+C
axol replay-dataset --repo_id myorg/pick-place --episode 3 --fps 30 \
    --robot_config.axol_config.left_stiffness 1.0 --robot_config.axol_config.right_stiffness 1.0

Replay drives the arms open-loop from the recorded actions — it does not watch the scene (though the opt-in --teleop_torque_threshold contact stop above halts playback and goes limp if a joint pushes hard against something unexpected). Make sure the workspace is clear and matches roughly what the episode was recorded in before starting, and keep the Stop control (or Ctrl+C) within reach. Match the stiffness used at collection time so the arm tracks the same way it did when recorded.