BEHAVIOR-1K (BEHAVIOR Challenge)

Stanford's simulated benchmark of 1,000 everyday household tasks, built on the OmniGibson simulator.

BEHAVIOR-1K is an embodied-AI benchmark from the Stanford Vision and Learning Lab, guided by Fei-Fei Li, Jiajun Wu, and others, with the first version published at CoRL 2022. Its tasks come from a survey asking what people would want a robot to help with around the house, distilled into 1,000 everyday activities spread across 50 interactive scenes (homes, gardens, restaurants, offices, and more), with more than 9,000 objects annotated with physical and semantic properties; task goals are defined in BDDL, a language that describes goal states with logical predicates. Its companion simulator, OmniGibson, is built on NVIDIA Isaac Sim and supports rigid bodies, soft bodies, and liquids. The team has run the BEHAVIOR Challenge since 2025; the second edition, in 2026, requires completing 100 full household tasks, provides 20,000 teleoperated demonstrations totaling 1,950 hours, and ranks entries by an average task success score that awards partial credit.

ExampleThe 2026 BEHAVIOR Challenge provides π0.5 and GR00T N1.7 as baselines; entrant policies control the robot using RGB, depth, and proprioceptive input to do household chores, with submissions due October 16, 2026, and a total prize pool of $11,000.

Also called

BEHAVIOR, BEHAVIOR Challenge

Related

OmniGibson · NVIDIA Isaac Sim · Household Tasks · Long-horizon Task · Benchmark · Mobile Manipulation

Sources

BEHAVIOR 官网 (Chinese)
BEHAVIOR-1K: A Human-Centered, Embodied AI Benchmark with 1,000 Everyday Activities and Realistic Simulation (arXiv 2403.09227)
2026 BEHAVIOR Challenge

As of

2026-09