Motor intelligence in a simple distributed control system: Walking machines and stick insects

Walking in insects and machines is considered as an example of a difficult motor control problem involving many degrees of freedom and strong constraints by the environment. A control system based on experimental results from stick insects and other arthropods is proposed; results from simulations and initial tests in a walking machine are presented. The control system relies on decentralized control at three levels ranging from individual joints to interleg coordination, and including intraleg step pattern generation and interleg coordination of step timing and leg forces. This decentralized control immensely reduces the demands on central supervisory systems. Control modules and mechanisms are formulated in neural terms and optimized using connectionist learning rules and genetic algorithms. Several approximate algorithms suggested by biological results are used in place of classical, explicit solutions. The control system also relies heavily on the loop through the environment to simplify calculations. In particular, positive feedback at the level of individual joints is proposed as a way of exploiting the physical constraints to simplify the control problem. © 1997 Elsevier B.V. All rights reserved.