Authors: Adrian Boeing; Jacky Baltes; Thomas Bräunl
Addresses: School of Electrical, Electronic and Computer Engineering, The University of Western Australia, 35 Stirling Highway, Crawley WA6009, Australia ' Department of Computer Science, University of Manitoba, Winnipeg, Manitoba, R3T 2N2, Canada ' School of Electrical, Electronic and Computer Engineering, The University of Western Australia, 35 Stirling Highway, Crawley WA6009, Australia
Abstract: Dynamic balancing is a hard problem for biped mobile robots, as is requires real-time processing of complex formulas from noisy sensors. In this paper we demonstrate how an offline modelling and simulation step can help to improve balancing, by providing a first approximation for a biped walking pattern. In order to achieve this goal, we design a physics model of the biped robot, using spline functions for all joint movements. Then we optimise the system parameters using genetic algorithms with the help of a rigid body simulation library. Transferring the evolved control system from the simulation to the physical world poses a number of challenges, especially because of complex sensor and actuator noise in the real world and inaccuracies in the physics model. Methods to minimise these problems are the injection of artificial noise into the simulation process and/or the use of multiple dynamic simulation systems simultaneously. The larger variance in the resulting simulation will result in a generally smaller reality gap and subsequently a control algorithm that is more easily adapted to a real robot.
Keywords: dynamic balancing; biped robots; simulation; spline functions; genetic algorithms; rigid body; modelling; mobile robots; walking robots; legged locomotion; robot control; artificial noise.
International Journal of Mechanisms and Robotic Systems, 2015 Vol.2 No.3/4, pp.276 - 294
Received: 29 Apr 2015
Accepted: 05 Sep 2015
Published online: 11 Jan 2016 *