Skip to content

Laika v2 walking demo — LeRobotDataset with joint states, action targets and a front camera. Synthetic placeholder data.

dataset30 fpsrobotlaika12 filesdigests all verified by the hublerobotlaikabipedlocomotiondemoadded

Use with an agent

The MCP client entry (.mcp.json for Claude Code; see MCP setup). A read key is enough to browse; nothing here starts a robot.

{
  "mcpServers": {
    "lucen": {
      "command": "lucen-mcp",
      "env": {
        "LUCEN_API_KEY": "lucen_sk_...",
        "LUCEN_API_URL": "https://api.mousemouse.ai"
      }
    }
  }
}

Or hand the agent the task in words — the prompt names the tools and the exact commands.

You have the MouseMouse MCP server (lucen-mcp) configured against https://api.mousemouse.ai.
Work with the dataset repo lucen/laika-demo on that hub.
1. Read episodes with the dataset tools (list_episodes, get_episode_series) rather than downloading parquet.
2. Pull it locally only if you need the files: lucen pull lucen/laika-demo ./laika-demo.

README

README.md

laika-demo

A small teleoperation dataset for Laika v2: 3 episodes, 180 frames at 30 fps, 12 joint positions and 12 position targets per frame.

Synthetic placeholder. These frames were generated by make seed, not recorded on a robot — the trajectories are phase-offset sinusoids. The layout is a real LeRobotDataset v2.1 tree, so tooling that reads the real thing reads this too. It is replaced by the recorded dataset when that lands in assets/sample-dataset/.

Layout

PathWhat
meta/info.jsonFeature schema, fps, splits, chunk/file templates
meta/episodes.jsonlOne row per episode: index, tasks, length
meta/tasks.jsonlNatural-language task strings
data/chunk-000/*.parquetFrame tables (state, action, timestamps, indices)
videos/chunk-000/observation.images.front/3 x h264 mp4, 96x64

Using it

lucen pull lucen/laika-demo
lucen train --dataset lucen://lucen/laika-demo --template act