OmniGibson

OmniGibson is a GPU-accelerated photorealistic simulation platform for household-level embodied AI, developed by Stanford Vision and Learning Lab (SVL). Built on NVIDIA Omniverse and Isaac Sim, OmniGibson provides high-fidelity physics simulation with support for rigid bodies, deformable objects, liquids, and articulated mechanisms. The platform powers the BEHAVIOR-1K benchmark, which includes 1,000 everyday household activities grounded in 50 realistic scenes with over 9,000 objects. OmniGibson supports multiple robots including the Franka Panda and Fetch, and provides a Python API for training embodied agents with reinforcement learning and imitation learning. OmniGibson represents the next generation of embodied AI simulation, combining the visual fidelity of NVIDIA Omniverse with the scalability of GPU-parallelized physics. It has become a critical platform for research in household manipulation, long-horizon task planning, and sim-to-real transfer.