Abstract

AI planners and agents increasingly select robot skills from machine-readable descriptions, such as “pick and place, 8 kg payload, ±0.1 mm repeatability”, and dispatch them to physical machines. A skill that a controller advertises says nothing about whether this robot can perform it now, with its current tool, payload, calibration, temperature, wear and safety state, within the required deadline and with the intended physical result. Attested firmware, an accepted command or a reported protective stop are equally weak as evidence of physical capability, actuation or zero motion.

This disclosure describes a vendor-neutral, non-actuating method for qualifying a robot skill against evidence:

  • Each skill is bound to an issuer and revision, and to a specific robot instance with its hardware, controller, tool and sensor versions.
  • Declared capability is kept separate from measured capability.
  • Task requirements in payload, reach, torque, speed and precision are tested against calibrated limits after subtracting measurement uncertainty.
  • Calibration version, freshness, scope, metrological traceability and coordinate-frame transforms are checked explicitly.
  • The state of the independent safety controller is read, never overridden, and its freshness is verified.
  • A command is admitted only if the end-to-end response fits within the deadline.
  • Dispatch, controller acceptance, actuator application and physical completion are recorded as separate evidence.
  • During execution, capability can only be downgraded, and the remaining work is re-checked.
  • When a skill moves to another robot, source calibration and safety state are not reused without evidence from the target robot.

Predicates are combined with three-valued logic under an explicit objective, giving CONSISTENT, INCONSISTENT or UNRESOLVED. UNRESOLVED never implies zero motion; the physical response stays with the separately validated safety architecture.

The disclosure provides:

  • a 48-stage, six-phase workflow;
  • six embodiments;
  • 70 formal relations;
  • 40 checkpoints;
  • a comparison with ROS 2 and MoveIt, OPC UA Robotics and Industrie 4.0 capability-skill models, ISO 9283, ISO/IEC 17025 and GUM, ISO 10218 and ISO 13849, IETF RATS, real-time schedulability and TSN, and VDA 5050;
  • a non-actuating evaluator in Python, JavaScript and Go with 79 shared synthetic test vectors, all passing (about 2.5 to 9 microseconds per in-memory evaluation).

It is not a safety certification and makes no SIL or PL claims. The laboratory is synthetic, self-authored and not peer reviewed.

A companion industry-relevance section aligns the method with public roadmaps:

  • NVIDIA: Isaac GR00T, Jetson Thor, Isaac Sim, Isaac Lab and Isaac ROS, and the Isaac GR00T Reference Humanoid Robot.
  • Google DeepMind: Gemini Robotics 1.5 and Gemini Robotics-ER 1.6.
  • Other global humanoid makers: Boston Dynamics, Apptronik, Tesla, Figure AI and Agility Robotics.
  • Chinese humanoid and embodied-AI makers: AgiBot, Unitree Robotics, UBTECH, XPeng, Xiaomi, Galbot, Leju and Fourier, plus Huawei Ascend compute.

Disclaimer: The industry and product-roadmap relevance is an AI-generated alignment based only on public announcements and market reports. It is not an endorsement, partnership, sponsorship, licence, technical validation, safety assessment or evaluation by any company or standards body named. It does not state or imply that any company uses, needs, has assessed or infringes this method or any patent application. All company, product and standard names are trademarks of their respective owners and are used only to describe technical relevance. Patent pending: Indian Patent Application No. 202631129178 (source disclosure on Zenodo, record 23793543). Related PCT application: PCT/IB2026/055615, filed 4 June 2026 and publicly disclosed 16 July 2026. Readers must read the original disclosures before drawing any conclusion. The inventor is open to strategic collaboration or acquihire discussions.

Keywords

AI safety, robotics, robot skills, capability qualification, humanoid robots, embodied AI, physical AI, calibration, measurement uncertainty, metrological traceability, coordinate transforms, functional safety, independent safety controller, protective stop, real-time deadlines, command admission, physical-effect verification, capability derating, skill transfer, sim-to-real, three-valued logic, ROS 2, OPC UA, NVIDIA Isaac GR00T, Jetson Thor, Gemini Robotics, AgiBot, Unitree, UBTECH, XPeng Iron

Creative Commons License

Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License

Download

Share

COinS