Gennert, Michael A.
The purpose of the project is to mimic human’s learning and motion mechanisms in order to create an adaptive walking gait on compliant humanoid robot - Atlas. The project applies neural controller theory based on Central Pattern Generators (CPG) to reduce a state (parameter) space from 100 states to an average of 10 states. The goal of the learning mechanism that utilizes unsupervised learning based on self-organizing maps and reward that adapts throughout the learning process is to find global optimal set of parameters for CPG. The learning mechanism also utilizes Covariance Matrix Adaptation – Evolutionary Strategies in order to converge to the parameter region that leads to stable walking gait (success region) quickly.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.