Testing the Optimality of Human ‘Walking-to-Look’ Performance

Humans are active, rather than passive, perceivers of visual information. In everyday life we often move our eyes, heads, or whole bodies in order to gain a better vantage point: perhaps providing a better angle or more visual detail than is available at the current position. The current study investigated this ‘walking-to-look’ behaviour. Influential accounts suggest that movement choice results from an optimisation of the balance between two factors: the energetic costs of movement, and the benefits it may bring (in this case, allowing participants a better view of the environment). Participants behaving optimally should maximise the expected gain of a movement. We tested whether participants' ‘walking to look’ behaviour was optimal in a task where they approached a screen in order to better make a visual discrimination (which of four quadrants contained a coherently oriented set of Gabor patches: these ranged in size so that performance increased steadily as they approached the screen). For each person we first measured across the range of distances from the screen (i) visual performance on the 4AFC discrimination task and (ii) reward, quantified using a monetary task. We then convolved these two functions to construct an expected gain function for each participant, and compared behaviour in the free-choice task with the optimal (maximal) point of this expected gain function. Walking-to-look performance was competent, but non-optimal, suggesting that small- and large-scale movements may operate according to different principles. The results are discussed with reference to other possible models of performance