OmniGibson

Stanford's Omniverse-based household simulator, the underlying engine behind the BEHAVIOR-1K benchmark.

OmniGibson is a household-environment simulation platform developed by Stanford's Vision and Learning Lab (SVL, including Fei-Fei Li), built on NVIDIA Omniverse and PhysX 5. It is the underlying simulator for the BEHAVIOR-1K household-task benchmark, replacing the earlier iGibson 2.0. Its physics coverage is broad: rigid bodies, deformable objects, cloth, fluids, and granular materials, plus thermal effects such as fire, steam, and smoke, and it tracks extended object states such as temperature, being cooked or burnt, being soaked, on/off, being cut, and being dirty. The paper notes that more than half of BEHAVIOR-1K's activities cannot be simulated at all without fluids, soft bodies, and cloth. Scenes are rendered with ray tracing or path tracing. BEHAVIOR-1K contains 1,000 everyday activities, 50 interactive scenes, and over 10,000 objects, and it is the platform behind the annual BEHAVIOR Challenge.

ExampleIn a cooking task where the robot has to put ingredients in a pot and heat them, OmniGibson continuously tracks the ingredients' temperature, marks their state as “cooked” once it crosses a threshold, and judges task completion based on states like this.

Also called

BEHAVIOR-1K Simulator

Related

BEHAVIOR-1K (BEHAVIOR Challenge) · iGibson · NVIDIA Isaac Sim · NVIDIA Omniverse · Household Tasks · Long-horizon Task

Sources

BEHAVIOR-1K: A Human-Centered, Embodied AI Benchmark with 1,000 Everyday Activities and Realistic Simulation (arXiv 2403.09227)
BEHAVIOR 项目主页 (Chinese)
StanfordVL/BEHAVIOR-1K (GitHub)

As of

2026-09