Mehta, Biren
Coauthors(s): Biren Mehta Biomedical Engineering, OHE 500, University of Southern California CA 90089-2520 Kawato Brain Dynamic Project (ERATO/JST), 2-2 Hikaridai, Seika-cho, Soraku-gun, 619-02 Kyoto Japan Stefan Schaal Computer Science and Neuroscience, HND-103, University of Southern California, Los Angeles, CA 90089-2520 Kawato Brain Dynamic Project (ERATO/JST), 2-2 Hikaridai, Seika-cho, Soraku-gun, 619-02
Kyoto Japan
University of Southern California
Biomedical Engineering and Neuroscience
Computational Learning and Motor Control Lab University of Southern California Hedco Neurosciences Building, HNB-103 3614 Watt Way Los Angeles, CA 90089-2520, USA

Visuomotor Control of an Unstable Task

Visuomotor control of an unstable system puts stringent constraints on the permissible delays of information processing as well as the applied control strategy. In this presentation, a typical representative of such a system, the task of balancing a pole on a finger, performed by human subjects, is investigated and compared against theoretical results and data collected from an anthropomorphic robot performing the same task. The analyses support the following conclusions: - Visuomotor delays in this task are about 220ms. - Humans use internal states to compensate for these delays. - There is no evidence for intermittent control, and the data is well modeled by a linear control strategy. - Either model-predictive control (indirect control) oor tapped delay-line control could explain the data, and both methods are indistinguishable in normal behavior. - Preliminary data from 'blank-out' experiments during "virtual" pole balancing on a computer suggest that humans employ a predictive forward model of the task to compensate for the delays. - Predictive control in this task cannot be implemented as a Smith predictor (Miall et al., 1993) since this method is provably unstable for tasks with unstable dynamics (e.g., posture).