~/innate-os/workspace/custom_skills/:
Episode files
Each episode is an HDF5 file recorded at 30 Hz, up to 60 seconds long.T is the number of timesteps.
The
metadata.json format is documented in Policy-Defined Skills.Documentation Index
Fetch the complete documentation index at: /llms.txt
Use this file to discover all available pages before exploring further.
Reference for the files a physical skill is made of
~/innate-os/workspace/custom_skills/:
pick_up_cup/
├── metadata.json # skill config, see Policy-Defined Skills
├── data/
│ ├── episode_0.h5 # one file per episode
│ └── episode_1.h5
└── <run_id>/ # added when a trained run downloads
├── act_policy_step_135000.pth # model checkpoint
└── dataset_stats.pt # normalization stats
T is the number of timesteps.
| Dataset | Shape | Contents |
|---|---|---|
action | (T, action_dim) | Leader arm commands. action_dim is 6, or 10 with base motion. |
qpos | (T, joints) | Robot arm joint positions |
qvel | (T, joints) | Robot arm joint velocities |
images/main_camera | (T, 480, 640, 3) | Head camera, RGB |
images/arm_camera | (T, 480, 640, 3) | Wrist camera, RGB |
cmd_vel | (T, 2) | Base linear and angular velocity. Only with base movement on. |
odom | (T, ...) | Odometry. Only with base movement on. |
metadata.json format is documented in Policy-Defined Skills.