Motion Planning And Stability Control For Table Tennis Hitting Operations Of A Humanoid Robot

Humanoid robots have similar characteristics in physical configuration with human beings, which is advantageous in helping humans with daily lives and work in artificial environments. Among the different capabilities of a humanoid robot, operation capability, especially the ability to operate in dynamic and unstructed environments, is an important premise to assist humans in tasks. Table tennis playing involves vision detection, trajectory planning, artificial intelligence and other technologies closely related with operation capabilities of a robot, and has been an active field for studies on robotic operations. This dissertation will take the table tennis hitting operations of a humanoid robot as an example, and presents studies on such key issues as determination and adjustment of task specifications, trjactory generation for hitting operation, and stability preservation. The research issues and results are as follows.First, according to the requirements for a humanoid robot to provide assistance in man-made daily environments and to hit the table tennis with coordinated whole-body movements, a humanoid robot model is given, and forward and iverse kinematics are also presented. Additionally, a control system for table tennis hitting application is designed.Second, a kind of mathematical description of task specifications for hitting operations in humanoid robot table tennis is presented, which can be applied to describe the task specifications for hitting or throwing and other operations when a humanoid robot is either standing in place or moving from place to palce. In addition, a task specification adjustment method based on learning is presented for error reduction in landing point control. This learning based method estimates the landing point error in advance using the incoming ball positions and velocities and the desired landing point, then causes some changes to the commanded emergent ball speed which will lead to adajustments in task specifications. This learning based task specification adjustment method contributes in reducing the errors between the actual and the desired landing points.Third, a kind of trajectory planning method for coordinated whole-body movement is presented in order to simultaneously acquire required task specifications and optimized stability. According to the task specifications on the time of hitting operation, the trunk position is optimized so that the projection of the mass center is located at the designated place, and the configuration of the operating arm is decided. In order to generate the whole-body trajectories which begins at the initial preparation stage and ends at the time of the hitting operation, segmented polynominal is used to generate the arm trajectory and spline is used to generate the trunk trajectory. Through optimization of the trunk waypoints, the optimized stability of the humanoid robot can be achived.Then, a kind of double constraints based stability control method during feet-ground contact is presented in order to preserve the balance and prevent the slip due to the external disturbances or model errors. After analyzing the inverted pendulum plus flywheel based dynamics model, a real-time slip indicator is established to demonstrate how the translational and rotational motions works in slip prevention, and a generalzed acceleration controller is designed. Through evaluating the state wether the robot will lose balance or slip, required acceleration increments are generated to restrict the slip indicator and ZMP within the desired range to maintain the stability.Finally, using the table tennis hitting operation as an example, the proposed methods for task specification adjustment, whole-body trajectory generation, and stability control are verified on a physical humanoid robot.