SAPIEN (SimulAted Part-based Interactive ENvironment)

A UC San Diego robot simulation platform specialized for interacting with articulated objects like drawers and cabinet doors.

SAPIEN, short for SimulAted Part-based Interactive ENvironment, is a robot simulation environment proposed in 2020 by Hao Su's group at UC San Diego together with collaborators including Stanford, with the paper appearing at CVPR 2020. Its physics runs on NVIDIA PhysX, its rendering is based on Vulkan, and later versions added ray-traced rendering; starting with version 3.0 it supports PhysX 5's GPU-parallel simulation. Its distinguishing feature is the PartNet-Mobility dataset released alongside the paper: a large collection of articulated objects with joint annotations — drawers that pull open, cabinet doors that swing open, faucets, and the like — well suited to research on part-level perception and interaction, such as a robot opening a door or pulling out a drawer. The widely used ManiSkill family of manipulation benchmarks is built on top of SAPIEN.

ExampleLoad a cabinet model from PartNet-Mobility into SAPIEN and have a robot arm learn to locate the handle, grasp it, and pull the cabinet door open.

Related

ManiSkill · PartNet-Mobility · Articulated Object · Articulated Object Manipulation · PhysX · Simulator

Sources

arXiv 2003.08515 - SAPIEN: A SimulAted Part-based Interactive ENvironment
GitHub - haosulab/SAPIEN

As of

2026-09