People need to predict an action before it occurs
Aliases: anticipatory cue · action preview · feedforward intent
What it is
A robot action preview supplies enough information before motion or state change for nearby people to predict what comes next. A useful preview may need direction, target, affected space, or requested cooperation rather than a generic announcement that motion will occur.
Why it happens
People extrapolate early cues through a forward model and organise their own movement around the predicted state. Without preview, perception, interpretation, and bodily response begin only after robot motion has consumed part of the safety margin. Preview reduces uncertainty, but an ambiguous or unreliable cue establishes a harmful prior. Cue-to-action correspondence matters more than stimulus quantity.
Studying it
Presence, modality, specificity, and reliability of previews can be manipulated while measuring goal-prediction accuracy, avoidance onset, coordination, startle, workload, and recovery from a false cue. Detectability and semantic comprehension are separate: seeing a light does not mean knowing the destination. Training and trial order must be reported because cue meaning is learned.
Where it stops holding
Tiny low-risk actions that affect nobody may not justify explicit preview, and constant signalling produces alarm fatigue. Preview cannot make motion beyond braking capacity safe. A single modality fails under noise, occlusion, or crowding. If the plan is unresolved, communicate alternatives or uncertainty rather than a false definite route.
Applying it
- Identify the minimum pre-action information a bystander needs: start, direction, occupied space, and required yielding.
- Map light, sound, orientation, or projection consistently to actions, including a perceptible cancellation when plans change.
- Ask untrained participants to predict the next action under realistic noise and occlusion. Validate prediction accuracy, avoidance lead time, and erroneous moves, not mere cue detection.