RoboBuddy IDE
A workspace for the curiousPhysics Preview

Build a workcell.
Program a robot.
See what happens.

Write Python. Build laboratory scenes. Explore robot motion in your browser—with optional help from an AI agent.

Launch OpenArm Quick start

No installation. No robot required. Simulation is not hardware validation.

Inside the workcell01 / OPENARM
OpenArm simulation with a custom instrument housing, hollow vessel and repeated rack components on an authored bench.
Actual simulator viewBuild · Program · Observe
OpenArm in an authored laboratory scene. Synthetic test geometry; not a calibrated reconstruction or hardware demonstration.
Python programmingMuJoCo physicsWebMCP collaborationOpen source Free to explore

Code meets the physical world

One workspace. Room to experiment.

Keep your program, scene and observations together. Every robot workspace makes its capabilities and limits visible.

01 / PROGRAM

Write a real program

Start with a robot program, edit Python and inspect telemetry. Save a local draft or export your workspace.

02 / SIMULATE

Observe the physics

Inspect motion and contacts in MuJoCo. Compare requested targets with observed results; a completed command alone is not proof of task success.

03 / CONSTRUCT

Build your workcell

Create OpenArm scenes with rotated parts, hollow profiles and validated convex meshes. Record assumptions and inspect the collision model.

04 / COLLABORATE

Work with an agent

Opt in to bounded WebMCP tools in a compatible host. Let an agent inspect, build and program while you retain Stop and access controls.

Photo-guided scene authoring

From a reference photo
to a scene you can inspect.

Give a laboratory photo to your multimodal agent. The agent interprets it and describes new equipment through the builder—not just a fixed list of equipment names.

Approximation stays visible.

The app validates the submitted geometry. It does not independently recognize image content, recover hidden mechanisms or calibrate dimensions from a photograph.

Follow the WebMCP workflow

  1. Describe and account for the scene

    List visible items, uncertain geometry and unresolved equipment. Keep estimated dimensions and layout adjustments explicit.

  2. Construct, preview, then apply

    Stage a bounded scene and check a disposable physics candidate. Only explicit application replaces the active workcell.

  3. Program and evaluate separately

    Define a supported task before grasping. Read contact, lift, carry, release and settled-placement evidence instead of assuming the command worked.

Robot workspaces

Choose your next experiment.

Six robot workspaces, each with a distinct set of supported experiments.

Bimanual manipulation Start here

OpenArm

Two-arm rigid-body manipulation, contact diagnostics and general laboratory scene authoring. Start here for scene building and the visitor guides.

Open OpenArm →

Robot arm

SO-101

A physical block-transfer benchmark with Python control and observed grasp, transport and release checks.

Open SO-101 →

Mobile platform

LeKiwi

A physical mobile-courier workspace. Read the selected task’s controls, source assumptions and evidence limits.

Open LeKiwi →

Experimental humanoid

Asimov 1

Actuator, whole-body dynamics and standing experiments. Bounded motion attempts are not a verified general walking or grasping capability.

Open Asimov →

Policy-driven contact

MicroDuck

Physical locomotion, ground-contact and ball-kick tasks with pinned policy/runtime assumptions. No physical roller mode is claimed.

Open MicroDuck →

Humanoid dynamics & poses

Unitree G1

Bounded physical dynamics and standing, plus a separate kinematic pose workspace. Walking and dexterous-hand control are not supplied.

Open Unitree →

Simulation parameters and authoring limits are not hardware measurements or payload ratings. No laboratory liquids, chemical processes or automatic collision-free motion planner are included in the general scene builder.

Start with an experiment

Your next idea starts
in the workcell.

Choose OpenArm, wait for Ready and explore the starter program. Take it further with your own scene or an agent-assisted experiment.

Launch the OpenArm workspace ↗

A desktop browser is recommended for editing. The manuals work on mobile. The IDE needs access to its model and runtime assets.

OpenArm manual →First run, Lab Builder, physical evidence and troubleshooting.WebMCP manual →Agent access, tool reference, photo workflow and runnable example files.