Acquisition and transfer of models of visuo-motor uncertainty in a throwing task
53.3037, Tuesday, 19-May, 8:30 am - 12:30 pm, Banyan Breezeway
Hang Zhang1,2, Mila Kulsa1, Laurence Maloney1,2; 1Department of Psychology, New York University, 2Center for Neural Science, New York University
We investigated how well people model their own visuo-motor error distribution in a throwing task and how well they transfer this model to a novel but predictable situation. Methods: The experiment consisted of three phases. Training phase. Subjects threw beanbags underhand towards targets displayed on a wall-mounted touch screen for 300 trials. The distribution of their endpoints was bivariate Gaussian. Choice phase. We used the 2-IFC task by Zhang, Daw, & Maloney (2013, PLoS Comp Biol): subjects repeatedly chose which of two targets varying in shape and size they would prefer to attempt to hit. Their choices allowed us to estimate their internal models of visuo-motor error distribution. Transfer phase. Subjects repeated the choice phase from a different vantage point, the same distance from the screen but with the throwing direction shifted 45 degrees. From the new position, visuo-motor error was effectively expanded horizontally by sqrt(2) and good performance required that subjects allow for this expansion in their judgments. Fifteen naïve subjects participated. For each subject, we estimated the horizontal and vertical standard deviations of her distribution models in the choice and transfer phases and compared them to those of her true error distribution. Results: (1) In their models for the choice phase, subjects underestimated the vertical-to-horizontal ratio of their true error distribution (mean 1.17 vs. 1.84), effectively assuming a more isotropic model. (2) Subjects’ models in the transfer phase had a vertical-to-horizontal ratio close to 1/sqrt(2), agreeing with an objectively correct transformation of their incorrect isotropic models in the choice phase. (3) The horizontal and vertical standard deviations in subjects’ models were highly correlated (Pearson’s r = 0.89 for choice and 0.92 for transfer), while the counterpart correlation for the true distribution was only 0.58, favoring that subjects’ distribution models were coded in polar coordinates.