DEVELOPER DOCS

Build against the behavior layer.

Behavior specifications, experiments, benchmark runs and adapters form the platform contract.

Behavior API

GET /api/behaviors
GET /api/behaviors/:id

Discover reusable task specifications.

Experiment API

POST /api/experiments/run
GET /api/experiments
GET /api/experiments/:id
GET /api/experiments/compare?ids=...

Run and inspect reproducible multi-seed experiments.

Authentication

Authorization: Bearer hb_live_...

Use account sessions or API keys for authenticated endpoints. Create and revoke keys from Workspace.

Benchmark request

POST /api/benchmark
Authorization: Bearer hb_live_...
Content-Type: application/json

{"behaviorId":"pick-place","completedSteps":[...]}

Returns a reproducible behavior evaluation and persists the run to the workspace.

Usage

GET /api/usage
GET /api/keys

Inspect plan limits, monthly benchmark usage and active API keys.

Execution boundary

simulation-harness
measured=false

MuJoCo
measured=true

The platform keeps deterministic harness results separate from measured simulator results.

Health & readiness

GET /api/health
GET /api/readiness

Use readiness checks in production deployment monitoring.

Robotics adapters

MuJoCo
Isaac Lab
Gazebo
ROS 2
HTTP robot gateway

These are planned execution contracts. The current public measured path is the supported MuJoCo worker; robot-specific adapters are not enabled yet.

Robot models

GET /api/robots
GET /api/robots/:id

Discover catalog metadata. A catalog entry does not imply that robot-specific evaluation is currently enabled.

Public experiment reports

POST /api/experiments/:id/share
GET  /api/share?token=...

Completed experiments can be shared as public read-only reports with reproducibility metadata and raw artifacts.

Reproducibility

behavior + version
engine + engine version
seeds + policies + duration
compute build + worker

Reports expose the exact evaluation configuration used to produce measured results and provide a one-click reproduction path.

Version comparison

POST /api/behaviors/compare/benchmark

{"behaviorId":"pick-place","versions":["1.0.0","1.1.0"],"seeds":[42,1337],"policies":["policy-a","policy-b"],"seconds":10}

Version comparison is currently disabled while the comparison worker is being hardened. Single-version MuJoCo evaluations are the supported measured path today.

QUICKSTART

Run your first benchmark

Create an API key in Workspace, then call the benchmark endpoint.

curl -X POST https://humanoidbehavior-com.4gmxsol.workers.dev/api/benchmark \
  -H "Authorization: Bearer hb_live_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{"behaviorId":"pick-place","completedSteps":["Perceive target object","Navigate to workspace","Reach and grasp","Transport object","Release at target","Verify placement"],"model":"my-policy"}'

Response

{
  "status": "passed",
  "behaviorId": "pick-place",
  "adapter": "spec-validator"
}