robosuite

A modular, MuJoCo-based robot-learning simulation framework and benchmark, maintained under the ARISE initiative.

robosuite is an open-source simulation framework developed by Stanford and UT Austin researchers under the ARISE initiative, built on MuJoCo, with its paper published in 2020. It breaks robots, grippers, controllers, scenes, and tasks into composable modules, offering both joint-space and Cartesian-space (directly controlling end-effector pose) controllers. It ships with built-in single-arm tasks such as lifting a cube, stacking blocks, opening a door, and wiping a table, plus three bimanual tasks, and it supports teleoperation for collecting demonstrations. Version 1.5, released in October 2024, added humanoid and other embodiments, whole-body controllers, photorealistic rendering, and sensor models. Both robomimic and RoboCasa are built on top of it, making it a common foundation for imitation-learning and reinforcement-learning research.

Examplerobomimic's standard datasets, such as Lift, Can, and Square, were all collected inside robosuite environments using a Franka Panda arm.

Also called

robosuite v1.5

Related

MuJoCo (Multi-Joint dynamics with Contact) · robomimic · RoboCasa · MimicGen · Simulator · SpaceMouse Teleoperation

Sources

robosuite 官网 (Chinese)
robosuite 文档：Environments (Chinese)
robosuite: A Modular Simulation Framework and Benchmark for Robot Learning (arXiv 2009.12293)

As of

2024-10