Abstract

“8 kg, ±0.1 mm” Is a Promise, Not a Proof: Evidence-Based Qualification of Robot Skills Before and During AI-Driven Execution


An AI planner reads “pick and place, 8 kg payload, ±0.1 mm repeatability” and dispatches the skill to a robot. But that advertised skill says nothing about whether this specific robot, with its current tool, payload, calibration, temperature, wear and safety state, can perform the task right now, within the required deadline, and with the intended physical result. An attested firmware image, an accepted command or a reported protective stop is equally weak evidence: it does not prove capability, it does not prove actuation, and it does not prove that the machine is not moving. As foundation-model planners and agents begin to drive physical machines, this gap between what a robot claims and what it can actually do becomes a direct safety and reliability risk.

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 generation-qualified robot instance covering hardware, controller, tool and sensor generations. Declared capability is kept separate from measured capability. Task requirements for payload, reach, torque, speed and precision are tested against calibrated limits after subtracting measurement uncertainty, and calibration generation, freshness, scope, metrological traceability and coordinate-frame transforms are each checked explicitly. The state of the independent safety controller is read, never overridden, and its freshness is verified. A command is admitted only if an end-to-end worst-case response bound fits within the deadline, and dispatch, controller acceptance, actuator application and physical completion are recorded as four separate pieces of evidence. During execution, capability can only be downgraded, never silently upgraded, with the remaining work requalified; when a skill is transferred to another robot, source calibration and safety state are never reused without target-specific evidence.

Predicates are combined with three-valued logic under an explicitly declared objective. A proven contradiction yields INCONSISTENT, complete evidence yields CONSISTENT, and missing evidence yields UNRESOLVED. UNRESOLVED never implies zero motion; the physical safety response remains with the separately validated safety architecture.

The document provides a 48-stage, six-phase engineering workflow with flowcharts, formal definitions, six detailed embodiments, 70 formal relations and 40 engineering checkpoints. It compares the method with ROS 2 and MoveIt, OPC UA Robotics and Industrie 4.0 capability-skill models, ISO 9283 performance testing, ISO/IEC 17025 and GUM metrology, ISO 10218 and ISO 13849 functional safety, IETF RATS attestation, real-time schedulability analysis and TSN, and fleet interfaces such as VDA 5050. A non-actuating reference evaluator in Python, JavaScript and Go passes 79 shared synthetic test vectors. It is not a safety certification and makes no SIL or PL claims; the laboratory is synthetic, self-authored and not peer reviewed.

This disclosure is based on the Mothership application PCT/IB2026/055615, filed 4 June 2026 and publicly disclosed 16 July 2026, which claims priority from PCT/IB2026/053385, filed April 2026 and publicly disclosed 4 June 2026. The author is open to strategic collaboration, licensing or acquihire discussions.

Keywords: robot skill qualification; AI planner to robot dispatch; foundation models for robotics; physical AI; capability-skill model; declared versus measured capability; robot calibration; measurement uncertainty; metrological traceability; coordinate frame transform; payload and torque limits; functional safety; independent safety controller; protective stop; worst-case response time; real-time deadline admission; actuation evidence; physical-effect verification; capability derating; skill transfer between robots; three-valued logic; ROS 2; MoveIt; OPC UA; VDA 5050; ISO 10218; ISO 13849.

Suggested CPC classes (for search): B25J 9/1674; B25J 9/1692; B25J 9/1661; B25J 9/1694; G05B 19/4155; G05B 23/0205; G05B 9/02; G06N 5/04.

Industry relevance: The problem addressed applies wherever AI planners, foundation models or software agents select skills for physical machines. That includes robotics AI platforms (for example NVIDIA and Google DeepMind), humanoid and general-purpose robots (for example Tesla, Figure, Boston Dynamics and Agility Robotics), industrial robot and automation suppliers (for example ABB, FANUC, KUKA, Yaskawa, Universal Robots and Siemens), and logistics robot fleets (for example Amazon Robotics). It is also relevant to compliance with ISO 10218, the EU Machinery Regulation and the EU AI Act. This industry-relevance statement is AI-generated and does not assert that any named company uses, implements or infringes the disclosed method. Readers are strongly advised to read the original PCT filings, including PCT/IB2026/055615 and PCT/IB2026/053385, before drawing any conclusion.


Creative Commons License

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

Download

Share

COinS