Efficient Path Planning for Highly Articulated Robots using Adaptive Forward Dynamics

We present an efficient path planning algorithm for highly articulated robots with many degrees of freedom (DoFs). Our approach takes into account physical and geometric constraints and formulates the problem as constrained dynamic simulation. We simulate the motion of the robot using a sub-linear time adaptive forward dynamics algorithm. We also present fast collision detection and response computation algorithms for articulated models. Our planner computes an initial path for the articulated robot and refines the path by performing bounded-error dynamic simulation to ensure non-penetration. In practice, our planner scales well with the number of DoFs for highly articulated robots. We demonstrate its performance on models with hundreds and thousands of DoFs.