Agile and Intelligent Control based on Geometric Motion Primitives for Robotic Arms

Human motion control systems are complex and highly refined, which have the properties of self-organizing, self-learning, and self-adapting. The excellent motion characteristics can provide biological inspirations to control strategies for manipulators. In this paper, we draw on the control theory based on biological motor primitives which can represent the fundamental geometric invariant properties. Then the trajec-tory planning based on motion primitives is employed, which can reduce the computational effort and greatly improve the performance of agile control. In this paper, we propose a modeling method for motion primitives based on geometric invariants, and construct three styles of boxing motion primitives including straight punches, uppercuts and swing punches. Then, a deep neural network-based approach for boxing primitives is presented, which can represent geometric primitives as neural network models. Finally, the experiments are carried out on simulated and real robotic arms.