Augmenting RRT*-planner with local trees for motion planning in complex dynamic environments

Collision free navigation in dynamic environments, where motion of moving obstacles is unknown, still presents a significant challenge. Sampling based algorithms are well known for their simplicity and are widely used in many real time motion planning problems. While many sampling based algorithms for dynamic environments exist, assumptions taken by these algorithms such as known trajectories of moving obstacles, make them unsuitable for motion planning in real-world problems. In this paper, we present RRT∗ based motion planning in unknown dynamic environments. Effectiveness of our idea is demonstrated in multiple simulations with more than 15 simultaneously moving obstacles placed in various environments.