Evolutionary Multi-objective Optimization of Mixed Neural Network Controller for Hexapod Robot Navigation Application

• Main Authors: Yan-Ming Chen, 陳彥銘
• Other Authors: Chia-Feng Juang
• Format: Others
• Language: en_US
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/76289976526363210887

碩士 === 國立中興大學 === 電機工程學系所 === 105 === This thesis proposes a mixed neural network (NN) for a hexapod robot locomotion control. The mixed NN consists of a seven-node fully connected recurrent NN (FCRNN) controller for straight forward walking and a two-node sensor-feedback NN (SFNN) controller for obstacle boundary following (OBF). The seven-node FCRNN locomotion controller controls each hip joint of the robot by independent signals to improve the walking performance. The two-node SFNN OBF controller uses a few parameters to realize the function of detecting an obstacle and walking along its boundary. A left-right symmetric structure is adopted to reduce training cost and NN model size in building a complete collision-avoidance controller. A training environment and multi-objective functions are designed to perform evolutionary parameter learning of the two controllers using the non-dominated sorting genetic algorithm-II (NSGA-II). Besides, a proportional-integral-derivative (PID)-based target searching (TS) controller is used to merge with the mixed NN collision-avoidance controller to realize the hexapod robot navigation. In the end, simulation results in training and test environments verify the effectiveness of the control methods proposed in this thesis.