Abstract
Foraging tasks like berry picking are extended visual search tasks with good berries as targets. Many real-world tasks have foraging structures (e.g. looking for military infrastructure in satellite images of Afghanistan). Back in the raspberry field, you would typically abandon search of one area before every berry was picked. Optimal foraging theory (Charnov, 1976) proposes that animals leave a patch when the yield from that patch drops below average yield. This calculation is modified by transit time to the next patch: stay longer if transit time is longer. We created an 8 × 8 field of patches. Each patch contained 8–64 “berries”. “Good” berries were probabilistically bigger and redder than bad berries (d′ = 2). Observers selected a patch to forage. They clicked to pick and received auditory feedback about the goodness of the clicked berry. They could leave for another patch whenever they liked. They tried to maximize points (hit = +1, false alarm = −1) in 10-minute blocks. Picking rate was varied (easy vs hard) by making it unnecessary or necessary move the cursor around obstacles and transit time could be slow or fast (=10× slow). In a patch, initial picks were about 85% successful across all four conditions. Observers abandoned the patch when success fell to about 70% when picking was hard. They persisted to about 50% when picking was easy. When transit speed was fast, observers picked the next patch on the basis of apparent density of resource no matter how far away. At slow transit speed, observers tended to move to nearest neighbors. Observers maintained a constant rate of berries/second within a ten-minute block. In each patch, the berries/second rate drops as the berries become scarce. Observers appear to leave the patch when the instantaneous rate drops below the average rate for the block – exactly as Optimal Foraging Theory predicts.