Motion Planning of a 14 DOF Biped Robot For Crossing an Irregular Shape Ditch Using a New Speed Function Incorporating Terrain Features and Biped Dynamics

Abstract

In this paper, a new method has been proposed to plan a path between an initial point and goal point on a terrain containing an irregular shape ditch. The path planner models this problem as a wavefront propagation in a Nonuniform medium. The Eikonal equation is a mathematical representation of this problem. To solve this equation a speed function value at each node of discretized terrain surface is defined by using the local geometry of terrain and the ditch in the vicinity of node and dynamic behavior of biped robot at that node. The solution of the eikonal equation gives a potential surface of the terrain. Gradient descent algorithm provides a path on this potential surface. A dataset which consists of a collection of energy optimal and dynamically balanced gaits (hip and foot trajectories) of a 14 DOF biped robot for walking on uneven terrain is created by using a genetic algorithm. This dataset is generalized by using a feedforward neural-network by training the network with stochastic conjugate gradient algorithm. The footstep planner sequentially plans the feasible footsteps along the obtained path with minimum deviation by geometric method. The trained neural network estimates the hip and foot trajectories for each step along the path so that the biped robot can follow the path by crossing the encountered ditch successfully in an energy-optimal way. The proposed method is validated by simulation of a 14 DOF biped robot for the walk between an initial point and goal point on a terrain containing an irregular shape ditch.