FAST GENERATION OF EFFICIENT MOTIONS FOR DIFFERENTIAL DRIVE ROBOTS

This paper presents a methodology for generating, in real-time, efficient trajectories for differential drive wheeled robots. Given a dynamic model of a differential drive robot, a reasonably large training set of minimum torque trajectories are first obtained for various final configurations. The trajectories are then organized into categories, and a principal component analysis performed for each category. The principal components are then used as basis functions for rapidly interpolating trajectories to arbitrary ending configurations. Through a case study we examine some of the issues underlying the numerical optimization and classification of trajectories, and report on the performance. The methodology is easily generalizable to other robotic platforms, as well as other training motion data and different physical notions of efficiency.