Tight Dynamic Movement Primitives for Complex Trajectory Generation

Abstract

Movement primitives are basic building blocks facilitating generation of complex trajectories. Dynamic movement primitives are second-order non-linear motion controllers. In this work, a new formulation termed Tight Dynamic Movement Primitives (TDMP) is introduced. The parameters of each TDMP are the demonstrated trajectory coordinates. During run-time, the stored trajectory is transformed, based on the actual start and target positions, and the motion controller parameters are calculated. An additional component driving the trajectory to the target can be added. The method was tested for trajectory generation of the ten digits and implemented in hardware for controlling a robotic manipulator. Results show that trajectories generated using TDMP are more similar to demonstrated paths than when using DMP. In comparison to the DMP fewer primitives are required for complex trajectory generation, although each primitive is encoded by more parameters. The additional TDMP run-time computation cost is small and the method remains suitable for run-time applications.