Real-time Kinodynamic Planning for Complex Systems

Increasingly complex robotic systems demand motion planning that can produce dynamically feasible trajectories while enforcing other difficult constraints. One particularly challenging example is microgravity robotic free-flyers, i.e., satellites with manipulator arms of non-negligible mass. These systems (when free-floating) are nonholonomic and often have high-dimensional state spaces. As such, trajectory optimization is a challenge: direct methods might fail to produce even a feasible solution. To account for the free-flyer dynamics we use a sampling-based method, kinodynamic-RRT, and explic- itly forward propagate the satellite dynamics under actuator constraints. This approach is meant as a way to produce feasible reference trajectories in the face of “difficult” dynamics to hand off to solvers that desire a feasible initialization trajectory, and is a stepping-stone toward future development of efficient optimizing sampling-based planners that might tackle this problem; solving the requisite two-point boundary value problem for optimizing sampling-based planning remains an open research challenge, and is a desired target of future investigation.

Future work will push at other real-time methods for dynamic systems.