Data & collection
Robot play data
Robot play data is interaction experience collected through exploratory control rather than a fixed sequence of labelled task demonstrations. It can preserve observations, actions, state changes, contact, and outcomes without requiring every segment to begin from a standard reset or complete a named task; its usefulness depends on how the learning objective interprets that off-task behaviour.
Also known as: play data, robot play, task-agnostic play data
Updated
Play records interaction without requiring a scripted task
Conventional task demonstrations are usually collected to show a particular successful behaviour from an intended start to an intended outcome. Robot play relaxes that structure. A person or exploration policy may press, move, grasp, release, retry, and transition between behaviours without segmenting every action into a named task.
Learning Latent Plans from Play describes teleoperated play as continuous interaction gathered without task segmentation, labelling, or repeated resetting. Its Play-LMP method learns reusable latent plans from that record. The term is not a promise that the data is random, autonomous, or label-free in every sense: actions, observations, timestamps, robot state, and sensor values can still be present.
Play and failed demonstrations are not synonyms
A failed task demonstration has an intended goal and an outcome that did not meet it. Play may have no downstream task instruction at all. Both can contain useful contact and recovery information, but their provenance differs.
That distinction affects supervision. A learner can use a failed attempt as negative or recovery evidence when its goal and outcome are known. A play segment can teach how actions change the world without saying which change should be preferred for a later instruction.
ForceWAM makes the separation explicit. Successful demonstrations train a task-compatible visual-wrench future model. Both demonstrations and play train an inverse-dynamics model that maps a realised visual-wrench future to the action that produced it. In the paper's experiments, adding play to that separate executor helps, while directly mixing play actions into two end-to-end baselines hurts.
Useful play still needs structure
Play data should retain the same technical context expected of a task demonstration: synchronized observations and actions, coordinate frames, units, controller mode, calibration, episode boundaries, safety stops, and the relationship between commands and measured response. It should also be marked as play rather than silently presented as successful task execution.
Coverage matters more than the label. Repeating one easy motion produces little evidence about unusual contact states. Useful play explores transitions that successful demonstrations under-sample while remaining inside the platform's approved operating and safety limits.
For dataset buyers, the diligence questions are concrete: who or what controlled exploration, which sensors and actions were logged, whether play came from the target embodiment, how much is continuous versus overlapping windows, which task data was kept separate, and which model objective can use the off-task interaction without imitating it as a goal.
Sources
Related terms
Data & collection
Demonstration
A demonstration is a recorded example of how an intended task or behaviour is performed, usually represented as a time-aligned sequence of observations, states and actions. For humanoid robot learning, demonstrations may come from teleoperation, kinaesthetic guidance, motion capture or autonomous experts and provide targets for imitation.
Models & learning
Imitation learning
Imitation learning is a family of methods that learns a policy from examples of expert behaviour rather than specifying every control rule by hand. In robotics, demonstrations pair observations or states with actions, trajectories or inferred objectives. Behaviour cloning is one imitation-learning method; interactive and inverse approaches address different supervision and distribution-shift problems.
Models & learning
Inverse dynamics
Inverse dynamics is the problem of determining the forces, torques, or actions associated with a specified motion or state transition. In model-based robotics it commonly computes joint forces or torques from position, velocity, and acceleration under a dynamics model; learned inverse-dynamics models instead infer actions from observed states or outcomes and may not expose physical parameters.
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.
Data & collection
Robot training data
Robot training data is recorded experience used to train, fine-tune, or adapt models for robot perception, prediction, planning, or control. It can include sensor observations, robot state, actions, task instructions, rewards or outcomes, demonstrations, failures, and embodiment metadata. Not every dataset contains every field, but their timing and physical meaning must be clear.