Biped Gait Planning And Balance Control Of Humanoid Robot

Humanoid robot becomes to be the hottest topic in robotic field. This paper introduced the development of humanoid robot in the world and focused on the subject about stable biped gait generating and balance control of humanoid robot "KONG", which is designed by Cyber System and Control Institute of Zhejiang University. In this paper, the offline simulation environment of humanoid robot biped gait generating was designed, an optimal perpendicular momentum compensation algorithm was advised, an observed based preview control of time-delay system for real-time biped gait generating was proposed, the problem of real-time attitude estimation of humanoid robot was solved and finally our robot was succeeded in walking at 0.8km/h.The main aspects of my research are listed below:1. The off-line simulation platform for our humanoid robot was structured. The calculation of kinematic/inverse kinematic, angular/linear momentum and ZMP based on linkage model of our humanoid was fulfilled. Algorithm for planning foot trajectory was advised. And under this simulation environment, we simulated our robot walking at 1.0km/h, provided valid data for the mechanic design and motor selection.2. The optimal perpendicular angular momentum compensation algorithm was proposed, which is a complementary method for biped gait generating with ZMP theory. The general idea of this algorithm is using upper-body twisting to compensate the perpendicular angular momentum caused by legged motion of lower-body. This algorithm is proved to be effective by simulation and experiments on the robot.3. The observer based preview control of a time-delay system for online gait generating was proposed. Based on the cart-table model for biped gait generating and the pre-knowledge of the time delay feature of our robot, the observer for a time-delay cart-table model was constructed through Kalman state observer with sensor information from six-axis force/torque sensor and accelerometer, the optimal drive that tracking the prescribed ZMP was obtained by preview control. With this algorithm, our robot was able to walk at 0.8km/h stably.4. The method for on-line attitude estimation was presented. According to the kinematic model of attitude kinematics, the time-discrete chastic system model was formed and the optimal attitude sate was fused by kalman filter from sensor information of six-axis force/torque sensor, three-axis accelerometer and three-axis gyroscope. The kinematics representation used made it possible to construct a linear algorithm that can be shown to give convergent estimates for this nonlinear problem. The state estimator was evaluated in experiments demonstrating how the estimate are long-term stable even in the presence of gyro drift.