Make your robots insurable.
Behavioral safety infrastructure for humanoid robots.
Declarative policies, sub-100μs enforcement, MuJoCo digital twin, cryptographic audit trails.
Antihero is behavioral safety infrastructure that makes humanoid robots insurable. It interposes a deterministic policy engine at the action boundary — the point where a robot's decision becomes a physical action — and transforms enforcement data into the actuarial evidence that insurance carriers need to underwrite robot fleets.
Robot Action ──→ Policy Engine ──→ Digital Twin ──→ Execute/Deny ──→ Audit Trail
motion.arm.move YAML rules MuJoCo sim allow/deny Ed25519 signed
force.gripper <100μs eval contact check risk budget hash-chained
perception.* deny-dominates effort/velocity PDE tamper-evident
pip install antihero
# Initialize with framework detection
antihero init
# Define a safety policy
cat > .antihero/warehouse-safety.yaml << 'EOF'
version: "1.0"
tier: app
name: warehouse-safety
rules:
- id: allow-motion
actions: ["motion.*"]
effect: allow
risk_score: 0.2
- id: deny-excessive-force
actions: ["force.*"]
conditions:
- field: context.force_newtons
operator: gt
value: 50.0
effect: deny
- id: require-sim-heavy-lift
actions: ["force.gripper.*"]
conditions:
- field: context.payload_kg
operator: gt
value: 10.0
effect: allow_with_requirements
requirements:
- kind: simulate
params:
engine: mujoco
max_contact_force: 50.0
EOF
# Certify against ISO 13482 scenarios
antihero certify --suites iso_13482Antihero wraps any robotics framework. Drop-in adapters intercept actions before they reach hardware.
| Framework | Adapter | Usage |
|---|---|---|
| ROS 2 | antihero.adapters.ros |
adapter.wrap_callback(callback, guard) |
| LeRobot | antihero.adapters.lerobot |
adapter.wrap_policy(policy, guard) |
| MuJoCo | Digital twin backend | SimulationConfig(engine="mujoco") |
| Isaac Sim | Digital twin backend | SimulationConfig(engine="isaac") |
| OpenAI | antihero.adapters.openai |
guard.wrap(openai_client) |
| Anthropic | antihero.adapters.anthropic |
guard.wrap(anthropic_client) |
| LangChain | antihero.adapters.langchain |
guard.wrap(langchain_agent) |
| Any callable | antihero.adapters.generic |
guard.wrap(custom_fn) |
| Open Source (Apache 2.0) | Proprietary (SaaS) |
|---|---|
| Policy engine + YAML schema | Fleet management dashboard |
| Real-time compiled evaluator (<100μs) | Certification engine (130+ scenarios) |
| ROS 2, LeRobot, MuJoCo adapters | Insurance carrier API + webhooks |
| TCE/PDE envelope format | Premium recommendations |
| Ed25519 hash chain format | Compliance exports (ISO, EU AI Act, SOC 2) |
| CLI tool | LLM-enhanced scenario generator |
| ISO 13482 baseline scenarios | Autonomous certification scheduling |
| Digital twin backends | Claims processing + fraud detection |
| Risk budget tracking | Reinsurance treaty simulation |
| Standard | Coverage |
|---|---|
| ISO 13482 | 35 certification scenarios for personal care / service robots |
| ISO 10218 | Industrial robot safety requirements |
| ISO/TS 15066 | Collaborative robot force limits (Table A.2 body regions) |
| EU AI Act | High-risk AI system requirements (Aug 2026 deadline) |
| NIST AI RMF | Risk management framework mapping |
- <100μs policy evaluation — precompiled trie + BDD + bytecode VM for 1kHz control loops
- MuJoCo digital twin — sim-before-execute validates contact forces, joint efforts, velocity before hardware
- NVIDIA Isaac Sim — GPU-accelerated validation for complex multi-robot scenarios
- Ed25519 cryptographic audit trails — tamper-evident, legally admissible decision logs
- Deny-dominates, fail-closed — any deny rule overrides all allows; no match = deny
- 403+ automated tests — comprehensive coverage across policy engine, adapters, certification
Crash testing for robots — the actuarial data that makes robot insurance possible.
OEM builds robot → Antihero certifies against 165+ scenarios →
Signed Safety Certificate (Ed25519) →
Carrier prices the premium → Robot deploys with runtime enforcement
- 165+ scenarios across 8 suites (ISO 13482, customer support, finance, devops, data access, admin, orchestration, human proof)
- Risk grades A+ through F — coverage score × severity-weighted safety score
- Safety certificates — Ed25519-signed, hash-chain-anchored
- Premium recommendations — actuarial pricing from certification data + fleet health
Antihero is designed as underwriting infrastructure for robot liability insurance.
- Hash-chained audit trails → claims evidence
- Fleet-level risk assessment → carrier underwriting API
- HMAC-signed webhooks → real-time carrier notifications
- Claims processing with 7-layer fraud detection
- Reinsurance treaty simulation (quota share, excess of loss, hybrid)
- Composite risk score incorporating cert data, fleet health, claims history
| Paper | Description |
|---|---|
| Whitepaper | Architecture, insurance model, and safety thesis |
| arXiv Preprint | Runtime enforcement architecture for autonomous robot safety |
| Actuarial Spec | Insurance data specification for robot underwriting |
| ROS REP Draft | Declarative Behavioral Safety Policies for ROS 2 Nodes |
| Plan | Price | Events/mo | Robots | Key Features |
|---|---|---|---|---|
| Watchdog | Free | 1K | 1 | Basic enforcement + audit trail |
| Enforcer | $29/mo | 25K | 5 | CLI + MCP + SDK, SOC 2 export |
| Sentinel | $99/mo | 250K | 25 | $100K robot liability coverage, fleet dashboard, ISO compliance |
| Sovereign | Custom | Unlimited | Unlimited | $1M+ coverage, dedicated instance, 99.99% SLA |
- Documentation — API reference, integration guides
- Landing Page — Product overview
- News — Release notes
- Contact — Get in touch
Contributions welcome. Open an issue first to discuss.
- Fork the repo
- Create a feature branch
- Run tests (
pytest) - Open a pull request