Trajectory synthesis based on different fuzzy modeling network pruning algorithms

Abstract

In this paper, trajectory synthesis based on different fuzzy modeling network pruning algorithms is investigated. The learning scheme uses a fuzzy neural network as the controller. Simulations focus on refinement and a thorough understanding of the pruned fuzzy modeling network. The learning scheme can control a biped robot that satisfies prespecified constraints such as the steplength, crossing clearance, and walking speed. Different issues regarding the scheme have been examined. They include the effects of using weight truncation and the effects of using weight decay. The proposed scheme is tested with simulations of the BLR-G1 walking robot.