A neuromorphic learning strategy for the control of a one-leggedhopping machine

摘要

Summary form only given, as follows. An adaptive, neural networknstrategy is described for the control of a dynamic, locomotive system,nin particular a one-legged hopping robot. The control task is to makencorrections to the motion of the robot that serve to maintain a fixednlevel of energy (and minimize energy losses). While for many dynamicnsystems energy conservation may not be a key control criterion, leggednlocomotion is an energy intensive activity, implying that energynconservation is a primary issue in control considerations. The authorsneffect the control of the robot by the use of an artificial neuralnnetwork (ANN) with a continuous learning memory. Results are presentednin the form of computer simulations that demonstrate the ANN's abilitynto devise proper control signals that will develop a stable hoppingnstrategy using imprecise knowledge of the current state of the roboticnleg