StanfordVL/BEHAVIOR-1K resource

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1660 stars · 241 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

BEHAVIOR-1K is a simulation benchmark for embodied AI agents covering 1,000 everyday household activities across 50 interactive scenes, built on the OmniGibson simulator (NVIDIA Isaac Sim/Omniverse). The monorepo bundles the simulator, task definitions (BDDL), dataset, evaluation, and teleoperation tooling for training and evaluating agents on long-horizon mobile manipulation tasks.

Use cases

• benchmark embodied AI agents on household tasks
• train robots for mobile manipulation in simulation
• simulate realistic physics with fluids, cloth, and thermal states
• evaluate long-horizon task planning for home robots
• research human-centered everyday activities for robotics
• run the BEHAVIOR challenge evaluation pipeline

When to choose

• you need a realistic embodied AI benchmark grounded in real human needs
• you require rich physical simulation with fluids, deformables, and semantic object states
• you want 1,000 diverse long-horizon household tasks with reproducible evaluation
• you are targeting the annual BEHAVIOR challenge

When to avoid

• you need a lightweight or fast simulator rather than high realism
• you lack an NVIDIA RTX GPU or 32GB+ RAM
• you only need simple tabletop manipulation benchmarks
• you need a pip-installable or Docker deployment (currently unavailable)

Facets

dataset · maturity active

simulation benchmarking machine-learning robotics robotics artificial-intelligence simulation machine-learning windows python embodied-ai benchmark omnigibson household-tasks mobile-manipulation isaac-sim monorepo linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="StanfordVL/BEHAVIOR-1K")