Motion plan and control of humanoid walking robots

Abstract

Walking for most humans and animals is an easy task due to the inherent robustness and the natural dynamics of the walking mechanism. Walking, however, for humanoid robots is not that easy because of its nonlinearity, high dimensionality, and natural instability. Effective use of humanoid robots in unstructured environments for human beings requires that they have autonomous and reliable locomotion capability. Locomotion for humanoid robots can take many forms. This thesis covers motion plan and control of humanoid walking robots. Five consecutive stages are addressed from the perspective of stable dynamic walking. Firstly, a natural and efficient walking pattern is designed on the basis of the insight gained from human walking. The walking pattern involves the configuration with stretched knees. Secondly, in view of the modeling error, as well as the environment uncertainty such as the unevenness and inclination of the surface, a posture controller, which online controls the orientation of the upper body of the robot is developed from the viewpoint of stability. Walking stability of the robot is improved through the control scheme. In the third stage, a gait controller, which online controls the swing-leg is developed. Fast and precise trajectory tracking of the swing-leg is achieved which enables humanoid robots to quickly swing the swing-leg so that fast walking can be realized. In the fourth stage, a controller with the purpose of decreasing the landing force and improving the walking stability by means of force control is developed. In the final stage, strategies for restoring equilibrium in the presence of external disturbances to maintain upright standing posture are analyzed. Simple models are introduced to exploit boundaries that determine the strategies used for preventing a fall. Simulation and experiments were performed with the humanoid robot NAO developed at Aldebaran Robotics in France. Using the walking system, NAO achieved dynamic stable walking and reached a maximum walking speed of 0.24 m/s in experiments. Besides, with the presented walking pattern NAO could walk stably with almost stretched knees at a lower speed, e.g. 0.10 m/s, so that the corresponding energy consumption was lower since the robot did not bend the knees all the time, which means that the robot is able to walk for a longer time without battery charging and has less heating problem.