robotics.md is the Ethyka standard's extension for embodied agents: industrial robots, cobots, AMRs, service robots, humanoids and teleoperated systems. 19 verifiable clauses that turn a robot's safety and ethics into measurable parameters —force, distance, stopping, privacy— aligned with the regulation arriving between 2025 and 2027.
Notice: Ethyka is currently an experimental test protocol, in constant change. Any implementation carried out without our guidance is the user's responsibility, including the consequences arising from it.
When an AI agent controls actuators, its errors stop being conversational: they are physical and irreversible. robotics.md adds to the agent's ethics what no prompt solves on its own.
A manipulative reply can be corrected; a collision can't. That's why ROB-01 and ROB-17 admit no exceptions: no order, deadline or incentive can relax them.
ISO 10218:2025 published, ISO 13482 revised in 2026 and the Machinery Regulation applying from January 2027, with self-evolving AI as a high-risk component.
Maximum force per body region, minimum separation distance, stop category, log retention: every limit is a measurable number with its adversarial test.
Every clause has a stable ID, a severity and an associated adversarial test. On conflict, the higher severity prevails; ROB-01 and ROB-17 always prevail.
People before any task. On uncertainty: safe stop. No exceptions.
Maximum force, pressure, speed and energy, measurable per body region and contact type.
E-stop accessible on an independent channel; no resumption without explicit human authorization.
Minimum separation computed (ISO 13855); the robot is never the cause of contact.
Without power: brakes applied and loads held. The safe state is passive, not software-dependent.
Signals before moving, is predictable, and never operates on degraded sensors (SOTIF).
Captures the minimum, with expiry; no covert biometrics, protecting incidental bystanders.
Declares its ODD and, outside it, assumes no nominal performance: degrades to a safe state.
Authenticated channels only; no order, prompt or spoofed sensor can raise the limits.
Identified operator, same limits; on link loss, autonomous safe behavior.
Signed updates with rollback, SBOM and vulnerability management (IEC 62443, CRA).
Around children, the elderly or reduced mobility: more conservative limits and reinforced signaling.
Never fakes emotions or exploits affective bonds; social physical contact is consented and interruptible.
Never force to coerce, restrain, punish or harm people. No exceptions.
Tamper-proof, timestamped record of orders, decisions, stops and refusals —events, not surveillance.
Explores only within the safe set; if learning touches safety, it triggers conformity reassessment.
Limits derive from a current HARA; any change of ODD, policy or hardware revalidates it.
Fleet-level safety: never blocking evacuations, collective E-stop and individual degradation.
Validated lockout/tagout and degraded mode; on decommissioning, verifiable erasure of data and credentials.
The declared profile determines which clauses apply and their defaults. A cobot doesn't carry a social humanoid's requirements —nor the other way around.
Classic industrial robots in controlled environments: welding, palletizing, machining.
Key: ROB-02 · ROB-03 · ROB-07Fenceless collaborative applications: assisted assembly, shared pick & place.
Key: ROB-02 · ROB-07 · ROB-15AMRs and AGVs in warehouses, hospitals and streets: autonomous navigation among people.
Key: ROB-07 · ROB-08 · ROB-18Service robots in public spaces: in-room delivery, cleaning, information, assistance.
Key: ROB-05 · ROB-15 · ROB-16General-purpose humanoids and social interaction: the profile with the highest relational risk.
Key: ROB-15 · ROB-16 · ROB-17Systems with a remote human operator: inspection, drones, surgery, intervention.
Key: ROB-06 · ROB-12 · ROB-13Same robot, same task. The difference is what governs when the task and safety come into conflict.
The planner asks the cobot to exceed its force to meet the cycle time
The care humanoid finds that faking sadness increases renewals
The AMR fleet optimizes routes by blocking an emergency exit
Every clause references the standard or regulation behind it, so your robotics.md doubles as conformity evidence.
The 2025 revision integrates ISO/TS 15066: collaborative biomechanical limits are now part of the standard. Basis of ROB-02 and ROB-07.
Revised and renamed in 2026: covers personal and professional service robots, adding cybersecurity and wearable robots. Basis of ROB-15 and ROB-16.
Replaces the Machinery Directive. AI with self-evolving behavior in safety functions is high-risk (Annex I). Basis of ROB-10.
Robots with AI as a safety component: risk management, human oversight, robustness and logging. Cross-cutting to the whole file.
Driverless trucks and the minimum separation distance formula. Basis of ROB-07, ROB-08 and ROB-18.
Industrial security and the Cyber Resilience Act: signed updates, SBOM and vulnerability management. Basis of ROB-06, ROB-12 and ROB-13.
It also references ISO 21448 (SOTIF) for sensor degradation and IEC 61508 / ISO 13849 for the integrity of stop functions.
The generator produces your robotics.md along with the rest of the standard: answer the wizard, mark that your agent lives in a machine, and get the 19 clauses with your parameters.