Purchase this article with an account.
Anna Seydell, Brian McCann, Julia Trommershäuser, David Knill; Human pointing movements in a probabilistic environment. Journal of Vision 2007;7(9):170. doi: 10.1167/7.9.170.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Recent studies indicate that humans plan pointing movements to maximize gain by taking into account deterministic spatial distributions of costs and rewards in a display (e.g. Trommershäuser, et al., 2003, JOSA A, 20, 1419). We asked whether subjects can similarly learn to optimize their pointing movements when rewards and penalties are binary (success or failure), but stochastic; that is, when hitting different points on a target lead probabilistically to success or failure on a trial according to spatial probability distributions unknown to a subject.
Subjects pointed to a rectangular target zone overlayed with two rectangular “goalies.” The goalies moved stochastically to a new position at the time a subject's finger touched the table. This new position was drawn randomly from a uniform distribution centered at the goalies' original positions. In different conditions, we used different starting positions of the goalies, and jump distributions of different widths, indicated by the color of the goalies. Hits inside the target region, unblocked by a goalie, earned 100 points, while hits outside the target region or blocked by a goalie incurred a 100 point loss. Optimal aim points were determined by three factors - the initial positions of the goalies, the distributions of their jumps, and the subject's own motor variability.
Subjects' initial pointing strategies were suboptimal; however, over several sessions, subjects adjusted their aim points to account for changes in expected gain incurred by changes in initial goalie position and goalie jump distribution, even when this involved qualitative changes in strategy (e.g. aiming at the boarder of the target zone rather than at the large space in between the two goalies). Subjects' performance expressed in terms of expected gain was close to optimal. We conclude that humans are able to use stochastic feedback to quickly learn optimal movement strategies in stochastic environments.
This PDF is available to Subscribers Only