LeIsaac

Lightwheel's open-source Isaac Lab teleoperation data-collection framework, plugging the SO-101 arm into the LeRobot pipeline.

LeIsaac is open-sourced by Lightwheel (光轮智能) under the Apache 2.0 license. It places a simulated SO-101 robot arm (single- or dual-arm) inside NVIDIA Isaac Lab, letting a user teleoperate the simulated follower arm with a real SO-101 leader arm, with keyboard and gamepad also supported; besides manual teleoperation, it can also generate demonstration trajectories automatically with a state-machine script. The collected HDF5 data can be converted to LeRobot's dataset format, used to fine-tune policies such as GR00T N1.5/N1.6 and deploy them to a real robot. It's also the official simulated imitation-learning environment for LeRobot's EnvHub, loadable in one line of code for tasks such as picking oranges, lifting blocks, tidying a toy table, and bimanual cloth folding, and it can also run in the cloud when no local GPU is available.

ExampleUsing LeRobot's make_env to load the so101_pick_orange task from LightwheelAI/leisaac_env: three oranges are placed onto a plate, and the arm then returns to its initial pose.

Also called

leisaac

Related

NVIDIA Isaac Lab · LeRobot · LeRobot EnvHub · SO-100 / SO-101 Arm · Lightwheel · NVIDIA Isaac GR00T N1

Sources

LightwheelAI/leisaac (GitHub)
Hugging Face LeRobot Docs: LeIsaac × LeRobot EnvHub

As of

2026-09