Platform Core / GappAI Skills
Build once. Validate carefully. Adapt repeatedly.
Most robotics projects begin again from zero. The GappAI Skill Library is our long-term product direction for converting project-specific engineering into reusable Physical AI assets.
It is not presented as a catalogue of off-the-shelf products available today. Skills enter the library only as evidence, supported hardware and operating boundaries become clear.
02 / Proprietary system architecture
Three connected technology assets form GappAI’s intended IP core. They connect human expertise to configuration-bound robot capability — without making one manufacturer the center of the system.
Connects tasks, subtasks, objects, tools, decisions, movements and contacts with success criteria, failures and recovery paths.
Output: an inspectable work model and a foundation for training-ready data.
The Skill Core holds task logic, learned policies and shared interfaces. Robot Adapters connect it to each platform’s kinematics, sensors, tools and control system.
Output: reusable core plus configured robot interface. Every transfer requires its own adaptation and validation.
The versioned Skill Passport binds a skill version to its tested robot, sensor and environment configuration.
Output: a technical release evidence record. Not an independent safety certificate.
Permissioned operational data and human corrections return to the learning process. New versions pass validation again; no unvalidated learning in live operations.
This specification defines the target architecture, not an already validated software stack or universal robot compatibility. Simulation supplements the four capture sources; real-world testing remains necessary.
Explore release and operations
Task-relevant vision, motion, sequence and interaction data become the foundation for measurable robot learning.
A useful robot skill is more than a model file. It combines learned behavior with hardware requirements, operating limits, evaluation evidence and the components needed for safe adaptation.
Task-relevant vision, depth, touch and contextual understanding.
Manipulation, motion and control behavior for the defined operation.
Supported embodiments, payload, reach, degrees of freedom and compute.
Required cameras, force sensing, tactile sensing and calibration.
End-effectors, fixtures, task tools and physical interfaces.
Operating boundaries, stop and recovery behavior, human escalation and exclusions.
Completion, reliability, cycle time, intervention and safety criteria.
Robot-specific and environment-specific parameters and integration logic.
Simulation, test scenarios, integration instructions, monitoring hooks and support procedures.
Validated conditions, limitations, changes and supported configurations.
Each level communicates where a skill has been tested and what may still require engineering. A skill does not advance through language—it advances through evidence.
Evaluated in research, simulation or laboratory conditions. Not represented as field-ready.
Evaluated in a bounded customer-relevant environment against agreed criteria.
Demonstrated in real operations within documented conditions and limitations.
Validated on more than one supported robot platform or embodiment.
Reusability does not mean identical behavior everywhere. The core skill is combined with robot, environment, safety and operational layers for each deployment.
Hardware-awareDifferent robots require different kinematics, payload limits, sensors and control interfaces.
Environment-awareLighting, materials, layout, people and operating rules change deployment behavior.
Evidence-boundA validation result applies only to the documented configuration and conditions.
The following entries illustrate the task classes GappAI intends to develop through paid pilots and partnerships. They are product directions, not claims of currently licensed or deployment-validated products.
First development family: Industrial Kitting & Machine Tending. A bounded industrial starting point. The remaining entries are later research directions, not parallel product launches.
Identify, grasp, sort and stage specified parts; verify kit completeness and quality.
Target: test-cell validation, then a bounded pilotLoad and unload parts, respect machine handshakes and handle misplacement within agreed operating limits.
Target: test-cell validation, then a bounded pilotSurface perception, tool control, obstacle awareness and repeatable cleaning coverage.
Target evidence: pilot validationDeformable-object perception, grasping, folding, placement and quality checks.
Target evidence: research to pilotComponent identification, safe tool use, inspection and structured maintenance procedures.
Target evidence: research to pilotRoutine visual checks, exception capture, navigation and structured reporting.
Target evidence: researchOpen-world object handling, placement, routine learning and human-aware operation.
Target evidence: long-term researchData & IP / Defined boundaries
Technology can scale without treating confidential customer work as an unrestricted training asset. The architecture separates customer data, customer-specific adaptation and reusable platform technology.
GappAI Capture → GappAI Skills
The judgement, sequences and practical knowledge of experienced people are a starting point for robot learning. Let’s identify the expertise worth capturing—and the task worth engineering.
Discuss skill development
AI & Robotics EngineeringGappAI GmbH · BerlinCustomers do not buy an unsupported generic capability. Each commercial use is tied to a documented skill version, robot configuration, environment, safety envelope and validation status.
A paid pilot combines operation analysis, demonstrations, data, engineering and controlled validation to create the first evidence-bound capability.
A compatible skill is licensed, adapted to the robot, calibrated to the environment and revalidated for the customer's conditions before deployment.
Monitoring, approved operational feedback, recovery evidence and controlled releases support the skill after deployment and create the path to wider reuse.
Explore deployment and operationsPhysical AI Pilot / Skill development