Frequently asked questions
Answers about robot behavior benchmarking, reproducible evaluation and the HumanoidBehavior workspace.
Last updated November 18, 2026 · HumanoidBehavior
What is HumanoidBehavior?
HumanoidBehavior is a platform for evaluating humanoid robot behavior with reproducible, measured simulation. Teams define a behavior, run it against a policy with explicit seeds, and keep the configuration and results together as evidence.
What does a robot behavior benchmark measure?
A benchmark runs a defined behavior under stated conditions and reports task-relevant outcomes, including step completion and measured metrics. Interpret results alongside the task version, engine, seeds, policy and measurement configuration.
Do I need an account to run a benchmark?
Yes. Sign in or create a free workspace before starting a measured benchmark. The free workspace currently includes 25 benchmark runs per month and 5 private behaviors.
Which simulation engines are supported?
The current public measured path is the supported MuJoCo worker. Isaac Lab, Gazebo, ROS 2 and HTTP robot gateway adapters are planned execution contracts and are not enabled yet.
What is a behavior version, and why does it matter?
A behavior version identifies one exact task specification. Comparing policies is only meaningful when runs share the same behavior version, because a changed task definition can explain a changed result.
Can I compare two policies or two behavior versions?
Version comparison is currently disabled while the comparison worker is being hardened. Single-version MuJoCo evaluations are the supported measured path today.
Can I share an experiment report?
Yes. Completed experiments can be shared as public read-only reports with reproducibility metadata and raw artifacts, so another team can inspect the exact configuration behind a result.
Does a MuJoCo benchmark prove a robot is ready for hardware?
No. A simulation result is evidence under the stated simulated conditions, not proof of real-world hardware readiness. Hardware deployment requires separate validation, safety analysis and testing.