Adaptive dynamic balance of a biped robot using neural networks

Abstract

An adaptive dynamic balance scheme was implemented and tested on an experimental biped. The control scheme used pre-planned but adaptive motion sequences. CMAC neural networks were responsible for the adaptive control of side-to-side and front-to-back balance, as well as for maintaining good foot contact. Qualitative and quantitative test results show that the biped performance improved with neural network training. The biped is able to start and stop on demand, and to walk with continuous motion on flat surfaces at a rate of up to 100 steps per minute, with up to 6 cm long step.