Towards Dynamic Manipulation on an Anthropomorphic Robotic Table Tennis Platform

Abstract

Specialized robots whose morphologies address narrow tasks are capable of super-human precision, speed, and accuracy. However, with generalized anthropomorphic designs coming to the forefront of robotics, we have yet to achieve parity with the best human performance on these platforms, particularly in dynamic manipulation which encompasses tasks such as throwing, catching, and striking. This thesis documents work towards an enabled robotic table tennis platform that will allow the development of planning and control algorithms for dynamic manipulation. Specifically, a fully integrated hardware platform is presented with two candidate vision systems and a 5 DOF anthropomorphic robotic arm. A dynamics model of the ball is introduced and validated for predicting the trajectory of the ball. To strike the ball, a nonlinear trajectory optimization problem is formulated for the arm and shown to be capable of generating various types of swings. This formulation is applied to the lower 4 DOF case by additionally considering the timing of the strike. Finally, nominal ball striking is demonstrated on hardware for the case of planar ball trajectories.