Passive Dynamic Walking with Neuroevolutionary Controlled Torso

Abstract

Passive dynamic walking is the simplest and the most efficient form of biped locomotion. Albeit the simplicity, the walking gait of passive dynamic walkers resembles human walking in slops very well. The most important drawbacks of these robots is their sensitivity on initial conditions and the limited range of slopes they can have a stable walking cycle on. The aim of this work is to alleviate these problems by adding a link that resembles a torso and controlling it using a Neuroevolutionary algorithm. To achieve this goal, initially the dynamical equations governing this robot was extracted. Then the control algorithm is applied. The reason to use this control strategy is to mimic how natural evolution could play a role in robust biped walking. It was observed that this strategy improved both shortcomings associated with passive dynamic walkers and interestingly the robot shows torso movements familiar to that of humans when facing steep slopes or zero slope.