Continuous trajectory planning based on learning optimization in high dimensional input space for serial manipulators

Abstract

In order to generate trajectories continuously for serial manipulators with high dimensional degrees of freedom (DOFs) in a dynamic environment, a real-time trajectory planning method based on optimization and machine learning aimed at high dimensional inputs is presented. A learning optimization (LO) framework is established. Multiple criteria are defined to evaluate the performance quantitatively, and implementations with different sub-methods are discussed. In particular, a database generation method based on input space mapping is proposed for generating valid and representative samples. The methods presented are applied on a practical application—haptic interaction in virtual reality systems. The results show that the input space mapping method significantly elevates the efficiency and quality of database generation and consequently improves the performance of the LO. With the LO method, real-time trajectory generation with high dimensional inputs is achieved, which lays the foundation for robots with high dimensional DOFs to execute complex tasks in dynamic environments.