Unnatural motion triggers discomfort more readily than appearance does
Aliases: motion uncanny valley · motion incongruity
What it is
When a near-human robot shows stiff joints, incorrect acceleration or deceleration, or mistimed blinks or lip-sync, this motion–appearance mismatch exposes its artificiality more strongly than appearance falling short of realism on its own. Mori speculated that motion deepens the valley's floor, but "motion always triggers more strongly than appearance" is not a uniformly confirmed finding — it is a relative claim with more supporting studies than counterexamples, not a settled fact.
Why it happens
This connects directly to the mind-perception-mismatch account of the uncanny valley itself: static appearance only sets up an expectation, and motion is where that expectation is either fulfilled or broken. A specific candidate mechanism is predictive coding / prediction error — the brain forms expectations about the timing, acceleration profile, and joint coordination of biological motion based on appearance, and when motion violates those biological-motion regularities through stiffness, mistimed phase, or abrupt trajectory change, the resulting prediction error is amplified, producing discomfort stronger than appearance-only mismatch. Some fMRI work has also reported unusual activation in visual-motion prediction regions while viewing android motion, offered as indirect support for this account.
Studying it
A common design crosses appearance realism with motion naturalness factorially, using one motion-capture sequence and parametrically degraded variants, measuring eeriness, affinity, action-recognition accuracy, and gaze to see whether manipulating motion parameters produces a larger effect than manipulating appearance realism. It must be stated honestly that some studies have failed to find the predicted motion-amplification effect, so researchers should report the interaction term rather than comparing two isolated samples and assuming the effect holds.
Where it stops holding
This claim inherits the same evidentiary problems as the uncanny valley itself — fewer studies support "motion triggers more than appearance" than support the valley overall, so it cannot be treated as an independently confirmed conclusion. Motion quality is also entangled with speed, audio synchrony, task intent, and familiarity, and a valley measured from still images is not equivalent to a robot in actual motion. Hiding safety-related deceleration for the sake of smoothness is undesirable in industrial settings — safety-relevant motion should prioritize predictability over mimicking biological smoothness.
Applying it
When budget is limited, polish motion naturalness before pushing appearance realism further: align action intent, onset and offset timing, velocity change, gaze, and speech timing first, then consider raising skin and facial detail, and turn safety-mandated deceleration into a predictable rhythm rather than simply hiding it. Verify with a crossed comparison — one version holding appearance fixed while varying motion, another holding motion fixed while varying appearance — to see which produces the larger swing in affinity ratings, and retest at the robot's actual operating speed and realistic viewing distance rather than concluding from video or close-up static demos alone.