Multi-sensor fusion based on BPNN in quadruped ground classification

Abstract

Appropriate perception of different ground substrates plays an essential role in realizing adaptive quadruped locomotion. In this paper, we propose a multi-sensor fusion method based on Back Propagation Neural Network (BPNN) using in real-time ground substrate classification for adaptive quadruped walking. In order to collect the body gyro information, foot-ground contact force, Direct Current (DC) motor information and joint angle to train the network, we present the enhanced walk strategy with Center of Gravity (COG) adjustment method with 6-axis motion sensor feedback and realize steady walk gait on different ground substrates. Using these method, the quadruped robot Biodog realizes multi-sensor information collection while walking on six different ground substrates. Then we train the BPNN using the collected data after calculation and normalization. In network training, about 99.83% samples have been classified correctly using BPNN. In real-time testing, about 98.33% has been classified successfully using trained BPNN.