Randomized planning of dynamic motions avoiding forward singularities

Abstract

Forward singularities, also known as direct, or actuator singularities, cause many problems to the planning and control of robot motions. They yield position errors and rigidity losses of the robot, and generate unbounded actions in typical control laws. To circumvent these issues, this paper proposes a randomized kinodynamic planner for computing trajectories avoiding such singularities. Given initial and final states for the robot, the planner attempts to connect them by means of a dynamically-feasible, singularity-free trajectory that also respects the force limits of the actuators. The performance of the strategy is illustrated in simulation by means of a parallel robot performing a highly-dynamic task.