Models & learning
Robot skill library
A robot skill library is a versioned collection of reusable robot behaviours stored with the information needed to select, execute, and verify them, such as embodiment, preconditions, parameters, executable code or policy references, success checks, failure modes, and run evidence. A library supports composition and reuse but does not make a skill portable to a different robot or scene unless its assumptions still hold.
Also known as: skill library, robot behaviour library, robot behavior library
Updated
A library needs contracts, not only scripts
A stored motion file or controller script can reproduce behaviour under the conditions in which it was created. A skill becomes reusable when another planner, agent, or operator can determine when it applies, provide its parameters, execute it through a defined interface, and test whether it achieved the intended result.
The SkiROS2 platform organises robot programs as skills with preconditions, conditions that must hold during execution, and postconditions. That contract lets a task planner compose perception, manipulation, and external tools while reacting when the world no longer satisfies an assumption.
The RAX skill library adds evidence to a similar idea. Each task folder can contain the procedure, code, preconditions, checks, failure modes, run records, and camera frames. Its repository marks a skill as verified only after two unattended successful runs of the committed code. That threshold is a project rule, not a universal statistical standard, but it makes the meaning of the status inspectable.
Embodiment and environment remain part of the skill
A grasp procedure written for one hand inherits that hand's geometry, force limits, controller, cameras, calibration, and reachable workspace. A mobile-manipulation skill also depends on base shape, localization, collision margins, and the arrangement of the scene. These dependencies do not disappear because the behaviour has a stable name.
Useful metadata therefore names the robot embodiment, tool and sensor versions, coordinate frames, required objects, start conditions, allowed parameter ranges, safety limits, expected duration, success check, and known failure cases. Cross-embodiment reuse normally requires retargeting, new calibration, or a different implementation behind the same higher-level contract.
A skill library complements raw interaction data
A library is an index of capabilities, not a replacement for the runs that produced them. The execution evidence helps a reviewer distinguish a reusable behaviour from a one-off script and inspect whether failures or interventions were omitted.
Versioning matters when skills are composed. A longer task should record the exact revisions it called and the arguments passed to each one. If a reused skill is changed after a failure, the new evidence should remain linked to that revision rather than silently rewriting the history of earlier runs.
For dataset buyers, the practical question is whether the library preserves both sides of reuse: an executable artefact that another system can call, and enough synchronized observations, state, commands, outcomes, and provenance to decide whether that artefact is safe and relevant in a new setting.
Sources
Related terms
Hardware & control
Autonomy
Robot autonomy is the ability to perform an intended task from the robot’s current state and sensor information without human intervention during that task. Autonomy is contextual and can differ by function, environment and operating phase; it is not a single permanent capability level for the whole robot.
Models & learning
Policy
A policy is the decision rule that maps a robot’s current observations or estimated state, and sometimes a task instruction, to an action or probability distribution over actions. It can be hand-designed or learned from demonstrations, rewards or both. In humanoid robotics, its outputs may be joint targets, torques, end-effector changes or higher-level skills.
Hardware & control
Robot embodiment
A robot embodiment is the particular body and sensorimotor interface through which a robot perceives and acts. It includes morphology and kinematics, actuators, end effectors, sensors, physical limits, and the observation and action conventions exposed to a controller or learned policy. Two robots can perform the same task while having different embodiments.
Models & learning
Robot generalization
Robot generalization is a policy's ability to perform a learned behavior when test conditions differ from training, such as with new objects, environments, tasks, or robot embodiments. A useful claim specifies which conditions were held out.
Data & collection
Trajectory
A trajectory is a time-ordered sequence of states or observations, actions and, where applicable, rewards generated as an agent or robot evolves. A complete episode or policy rollout often yields a trajectory, but the terms are not universally identical: trajectories may be partial, while episodes have dataset- or environment-defined boundaries.