CNNs based Foothold Selection for Energy-Efficient Quadruped Locomotion over Rough Terrains

Abstract

When deployed in practical scenario, the legged robot has higher terrain passing ability but is suffering from lower locomotion efficiency than the wheeled robot. In this paper, we present a strategy that can improve the locomotion efficiency for a quadrupedal robot. First, an optimized energy-efficient nominal stance is generated. Second, a Convolutional Neural Networks (CNNs) based and self-supervised foothold classifier is implemented which will guide the robot to form the supporting legs in energy-efficient nominal stance during locomotion. The effectiveness of the present approach is validated on our quadrupedal robot Pegasus in stairs climbing experiment.