Inverse dynamics learned gait planner for a two-legged robot moving on uneven terrains using neural networks

Abstract

This paper proposes a method for design of inverse dynamics learned, neural network-based gait planner for a two-legged robot negotiating uneven terrains. The lower limbs' gaits are generated utilising inverse kinematics, and those of the trunk and swing foot are derived using a neural network aiming to maximise the dynamic balance margin. A genetic algorithm is used to provide training off-line to the gait planner. Its performance has been tested through computer simulations on different terrains, namely staircase, sloping surface and ditch. Simulation results show that the developed planner has successfully generated appropriate gaits to negotiate the uneven terrains.