Biped locomotion and stability : a practical approach

Biped locomotion has proven to be a very hard problem to solve. At current date, there exists no humanoid robot that can move as dynamic and robust as humans do. Even though there has been much research and certainly some interesting progress in biped locomotion, there is certainly no single solution that provides robots with the same capabilities as humans. What seems to be especially lacking is a quantitative method for comparing different solutions. In this work a general method for designing a locomotion controller, independent of the type of robot, is presented. Three distinct stages are identified: Designing the nominal gait, finding a proper encoding of the nominal gait and augmenting the nominal gait to provide stabilization. A nominal gait was successfully evolved using Particle Swarm Optimization algorithms. The nominal gait was encoded into a CPG Network which represents the controller as a dynamical system. Furthermore, a general framework of measuring stability, independent of robot and controller is defined. This allows quantitative comparison of controllers, feedback integration and gaits. Within this framework, stability of the nominal gait was compared with integration of feedback into the CPG. A significant improvement of stability under external perturbation was achieved using simple feedback.