Ioan Sucan, Kinodynamic motion planning for real robots

In addition to geometric constraints such as collision avoidance, real robots are limited in their motion due to bounds in acceleration, friction, gravity, etc. There are a variety of motion planners that can handle planning with dynamic constraints, but the obtained solutions usually require postprocessing in order to be executed on hardware. Newer motion planners use physics simulation to analyze the evolution of the robotic system in time. This produces results closer to the real behaviour of the robot, but is more computationally expensive. One new planner that attempts to maximize the ratio of covered space to simulation steps is presented in this work. Improvements in runtime of up to 3 orders of magnitude are observed.