01
Define
Represent a robot task as a reusable behavior with explicit steps, versions, compatibility and success criteria.
Browse behaviors →ROBOTICS EVALUATION INFRASTRUCTURE
Build robot behaviors.
Measure them with confidence.
HumanoidBehavior gives robotics teams a structured behavior layer and reproducible evaluation workflow — from task specification to measured MuJoCo results.
Platform ready·Browser MuJoCo compute·25 free benchmark runs / month·No card required
VERIFIED SAMPLE RUN10 measured runs
Benchmarkhumanoid-stand-v1
EngineMuJoCo
Seeds5 × 2 policies
EvidenceMeasured · persisted
A published sample run is shown as evidence of the measurement pipeline—not as a claim that every listed behavior has these metrics.
Inspect the raw experiment artifact →✦
Quick tipKeep the same seeds when comparing policies, then inspect the stored report.
Open Dashboard →LIVE EVALUATION WORKFLOWREADY
01BehaviorPick & Place
↓02EngineMuJoCo
↓03Evaluation5 reproducible seeds
↓04ReportSuccess · Cost · Contacts · Validation
TIP 01Start with 5 seeds
Keep the same seeds when comparing behavior versions so changes are easier to interpret.
TIP 02Read validation separately
A passed schema check is not the same thing as a measured physics result.
TIP 03Save the report
Each completed experiment keeps its configuration and measured output together.
StructuredVersioned behavior specifications
MeasuredPhysics-based MuJoCo evaluations
On-deviceRun supported benchmarks from your phone
ReproducibleExplicit seeds and stored runs
Developer-firstWorkspace + API workflow
THE PRODUCT
A behavior layer for robotics teams.
Stop rebuilding task definitions and evaluation scripts for every experiment. Start from a shared specification, run it, and keep the result with the exact configuration.
02
Evaluate
Launch controlled experiments with selected engines, seeds and policies. Measured MuJoCo runs stay separate from deterministic checks.
Open Experiment Lab →03
Inspect
Keep experiment configuration and result data together so teams can inspect validation, metrics and execution status.
Open workspace →MEASUREMENT, NOT MARKETING
Every evaluation has a trace.
Configure the run. Store the seeds. Execute the worker. Inspect the report. The platform is designed around reproducible evidence rather than static claims.
humanoidbehavior / experiment
$ run humanoid-stand-v1 --engine mujoco --seeds 42,1337,2026,7,99 --policies policy-a-stabilizer,policy-b-stabilizer
benchmark humanoid-stand-v1
engine MuJoCo
runs 10
status measured
→ measured survival, torso height, tilt and control cost
→ results persisted as a reproducible artifact
→ compare the two policies from the stored evidenceWHO IT'S FOR
Built around real robotics workflows.
Use the same behavior contract across research, policy development and evaluation pipelines.
Standardize evaluation.
Give engineers a shared place for behavior definitions, experiment history and benchmark configuration.
Make experiments reproducible.
Keep seeds, versions and engine choices attached to the result instead of scattered across scripts.
Measure policy iterations.
Use structured tasks and controlled simulation runs before moving a behavior toward hardware.
HUMANOID WORKFLOWS
Behavior patterns that map to humanoid work.
Start with reusable specifications for manipulation, whole-body interaction, navigation and human-robot handover.
01
Pick & Place
Manipulation sequence with explicit grasp, transport and placement steps.
View specification →02
Open a Door
Whole-body interaction pattern with alignment, force application and recovery steps.
View specification →03
Human Handover
Bimanual interaction pattern with recipient detection and transfer confirmation.
View specification →HOW IT WORKS
From behavior definition to evidence.
01Choose a behaviorStart with a reusable task specification.
02Configure the runSelect version, engine, seeds and evaluation settings.
03Run the workerQueue the evaluation and track progress.
04Inspect the reportReview measured metrics and validation state.
START AT $0
Your first benchmark is a few clicks away.
Create a free workspace, choose a behavior and launch a reproducible evaluation.
No credit card required · Free plan includes 25 benchmark runs/month