Foothold selection for quadruped robot based on learning from expert

Selecting the proper foothold is a key technology for the quadruped robot, although there has been so much progress in this area, most of the existing algorithms select foothold without considering the kinematic constraint of legs, and it is very tough to adjust the parameters of foothold selection model. In this paper, we are focus on designing the foothold selection model and learning its parameters from expert guiding. Firstly, the terrain features that influence the robot stability are detected; Secondly, foothold selection model is designed, which considers both the terrain features and the kinematic constraint; Thirdly, the model parameters are learned with the Support Vector Machine, and the training data is recorded in the simulation environment, including the terrain features of candidate foothold and their rank orders; Lastly, the effectiveness of the algorithm for computing terrain features is validated in simulation.