Use automatic pick and place to move the red cube into the green box, or control the six joints by hand. The default controller computes motion with inverse kinematics and staged trajectories. Everything runs in your browser; no robot is connected.
Loading the physics engine…
Drag to turn the view, scroll or pinch to zoom.
Open your own files
Open a folder with an MJCF model (.xml) and its meshes to load a robot of your own, STL files with the same names to swap parts of this one, or an ONNX policy to let it drive the servos. You can also drop files onto the 3D view.
Record up to 60 simulation seconds or 12,000 samples. Downloads contain initial state, controls, trajectory, versions and file hashes; local model files stay on this device.
SO-101 tasks record automatically. Starting another task replaces the previous episode; download it first to keep it.
Playback sets the stored poses. Rerun requires matching engine/model/policy hashes and recomputes physics from recorded controls, without policy inference. Results need not match bit for bit across environments. The pose book lasts until this page closes.
Record an experiment or save the current pose.
Servo target angles (degrees)
What this simulator shows
It runs the MuJoCo physics engine in your browser with the SO-101 arm model from MuJoCo Menagerie, made from TheRobotStudio's open SO-ARM100 design, on a table with four 3 cm cubes and an open box.
Each slider changes one servo target in this simulation. Joint names follow LeRobot, but simulated angles do not establish a real arm’s zero, direction, travel or calibration.
Automatic pick and place uses mathematical motion planning in the browser. You can save local experiment records; files are read locally. This page does not train a model or connect to a real arm, and physical behavior has not been validated.
Arm model: MuJoCo Menagerie robotstudio_so101 (Apache-2.0), based on TheRobotStudio SO-ARM100. Table scene after lerobot/so101-tabletop-mujoco (Apache-2.0). Physics by MuJoCo (Apache-2.0). Optional historical ACT policy: this neural policy was trained in the LeRobot SO-101 tabletop simulation (Apache-2.0). It reads six joint angles and scene positions rather than camera images. The archived report recorded 93/99 successes in its training simulator and 80/99 on those same scenes in the newer browser engine. These are historical ACT results, not measurements of the default mathematical controller or real-robot validation.