SAPIEN

SAPIEN (Simulation of Articulated Physical Interaction for Embodied AI) is a physically realistic simulation platform developed at UC San Diego for household robot tasks. It provides a rich environment with sufficient articulated objects, physically realistic simulation, and transferability to real robots, addressing the longstanding challenge of building home assistant robots. The platform distinguishes itself through its focus on articulated objects — doors, drawers, cabinets, refrigerators, and other household items with movable parts. SAPIEN provides physically accurate simulation of these articulated mechanisms, enabling robots to learn interaction skills that transfer to real-world counterparts. The simulator supports multiple physics backends and provides realistic rendering. SAPIEN is the foundational platform for the ManiSkill benchmark family. ManiSkill (SAPIEN Manipulation Skill Framework) and ManiSkill2 are built directly on SAPIEN, providing standardized manipulation tasks and demonstrations. The integration enables GPU-parallelized simulation for efficient reinforcement learning and imitation learning training pipelines. The platform supports a wide range of embodied AI research: task and motion planning, reinforcement learning, imitation learning from demonstrations, and visual navigation. With 784 GitHub stars, SAPIEN has an active research community and provides extensive documentation including tutorials, API references, and example workflows for new users. SAPIEN's key contribution to the embodied AI ecosystem is providing the infrastructure for Sim-to-Real transfer of manipulation skills on articulated objects. By modeling the kinematic and dynamic properties of real doors, drawers, and appliances, it enables researchers to train policies in simulation that can be deployed on real robots without the extensive engineering effort typically required for real-world manipulation.