Human Demonstration Trajectory Refinement for Redundant Manipulators

Abstract

Abstract Motion planning for robotic manipulators is a fundamental work in the research on brain-machine interface (BMI), as well as a key research issue in robotics community. Recently, algorithms for trajectory planning of end effector by learning from human demonstrations are reported, which could reduce the workload on coding for different manipulation actions. However, current approaches have at least one of the following drawbacks: (1) the generated trajectories usually are discontinuous; (2) physical limits of actuation system are not considered; (3) trajectory executive time is not optimal. This paper describes a novel method based on affine trajectory deformation to refine demonstration trajectory within manipulator's physical constraints and optimize the executive time. In the proposed method, the affine trajectory deformation is employed in joint space as an optimization problem to minimize the revision and time under the constraints on the physical limits of actuation system. The feasibility of our approach is demonstrated on the humanoid robot arm, which learned a table tennis strike and reproduced same motion as fast as possible.