(The) role of attention in the generation of human and robot arm movements

Abstract

Even as robots are being asked to perform increasingly complex tasks simultaneously, current robot motion planning and control laws for the most part fail to consider limitations on the available computation, communication, and memory resources. Humans on the other hand are adept at multi-tasking, continuously shifting their attention from one task to another. In this thesis we investigate the role of control attention in the general of both robot and human and arm movements. We examine robustness to spatiotemporal quantization as a means of evaluating the performance of arm trajectory generation and control laws from the perspective of control attention. We then develop a stochastic control strategy for generating and controlling arm movements that attempts to minimize the required computation and communication resources, based on the minimum intervention principle put forth by Todorov and Jordan. We perform numerical experiments for planar arm reaching motions with respect to these spatiotemporal quantization metrics. Experimental results show that our control laws share many of the distinguishing features of human reaching motions. We also examine the effectiveness of our measures in explaining human arm movement data. Specifically, we examine if the directional bias in arm movements reported recently can be explained with our attention-based principles for trajectory generation and control.