Skip to content
tetsugo02Public

About

Dora based robot experiment system

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

roboloom

English | 日本語 · Docs: English | 日本語

Tests Docs Python 3.12 Python 3.13 Python 3.14 Release License: MIT

Teleoperation, multimodal data collection, and policy inference on dora-rs, saving directly to LeRobotDataset (0.7.0).

Features

  • Process-isolated dataflow: Runs inputs, controllers, followers, sensors, policy, and recorder as separate dora-rs (1.0.1) nodes.
  • LeRobotDataset 0.7.0: Records synchronized robot state, actions, RGB cameras, DIGIT tactile sensors, and audio ready for lerobot-train.
  • Hardware safety & gravity compensation: Enforces per-joint speed clamping, jump rejection, EEPROM verification, watchdogs, and leader gravity compensation.
  • Interactive UI, CLI & Python API: Textual operator screen and launcher, --plain non-interactive CLI, and Session / InProcessSession Python APIs.
  • Extensible: Add custom robots, inputs, sensors, controllers, or policies via module:Class without modifying roboloom.
flowchart LR
  S["Session<br/>screen, CLI, Python"]
  I["input<br/>leader, Pika Sense"] --> C["controller"]
  P["policy"] --> C
  C --> F["follower<br/>Rakuda2, PiPER"]
  X["sensors<br/>RealSense, DIGIT"] --> R["recorder"]
  F --> R
  I --> R
  R --> D[("LeRobotDataset")]
  S -. "start, stop, save" .-> C
  S -.-> R
Loading

Hardware & Feature Status

Component Status Notes
Rakuda2 (leader teleop, gravity compensation, RealSense, DIGIT, custom policy & interventions) Hardware-verified Real hardware default
Local hooks (policy actions, policy teleop, lifecycle events, action.filtered) Mock only Explicit YAML write claims; existing speed and safety limits still apply
LeRobot policies (policy.lerobot, lerobot-train), replay, serve-policy, async inference Mock only Verified on mock dataflows; policy/train require Python 3.12–3.13
PiPER, Pika (Gripper, Sense, wrist cameras), webcam, audio In development Not verified on hardware; pika tracking requires Python 3.12 on Linux x86_64

See the full Support matrix and Status labels. Always verify joint limits, EEPROM, and stop behavior on your robot before hardware runs; never reuse mock limits on hardware.

Installation

Requires Python 3.12–3.14, uv, and Git. Add roboloom to your project:

uv init --python '>=3.12,<3.15' my-robot
cd my-robot
uv add git+https://github.com/tetsugo02/roboloom.git --tag v0.2.0 --extra rakuda2 --extra record --extra hardware --extra digit
Extra Purpose
(base) Core dora runtime, config validation, CLI, mock devices
rakuda2 Dynamixel SDK for Rakuda2 follower & Rakuda-mini leader
record LeRobotDataset recording (lerobot[dataset]==0.7.0)
hardware RealSense (pyrealsense2) and audio input (sounddevice)
digit DIGIT tactile sensors
piper, pika PiPER SDK, Pinocchio IK, and Pika tracker/cameras (in development)
policy, train LeRobot policy checkpoints and lerobot-train (Python 3.12–3.13)

When using record, policy, or train, add the NumPy override below to your project's pyproject.toml (required on Python 3.14; allows NumPy 2.3+ on 3.12–3.13):

[tool.uv]
override-dependencies = [
    "numpy>=2.0,<3; python_version < '3.14'",
    "numpy>=2.3.3,<3; python_version >= '3.14'",
]

roboloom also ships an Agent Skill for coding agents (uv run roboloom skills install). Prebuilt wheels and SHA256SUMS.txt for each release tag are attached on GitHub Releases. See Installation for details.

Quick Start

CLI

uv run roboloom init rakuda2          # generate .roboloom/rakuda2/
uv run roboloom validate --mock       # check config without opening devices
uv run roboloom run --mock            # launch the interactive operator screen

To record fixed-length episodes non-interactively:

uv run roboloom record --mock --task "handover" --seconds 10

See First run without devices (from a repository clone) or Rakuda2 setup for real hardware.

Python API

from roboloom import Session

with Session.launch("rakuda2", options=["gravity"], mock=True) as s:
    s.wait_ready()
    s.record("pick up the cup", seconds=20)

For setup/calibration helpers (roboloom.setup) and the single-process InProcessSession API, see the Python API reference.

Documentation

Full bilingual documentation is hosted on GitHub Pages (English / 日本語):

Development

git clone https://github.com/tetsugo02/roboloom.git
cd roboloom
uv sync --extra rakuda2 --extra test --group dev --group docs
uv run ruff check .
uv run mypy src
uv run pytest -m "not e2e"

See Contribution for adding robots, running E2E tests, and building the documentation.

License

Released under the MIT License. Bundled third-party robot models retain their own licenses (see src/roboloom/robots/piper/model/LICENSE.piper_ros).

About

Dora based robot experiment system

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages