Local Trajectory Planning Using Dynamic Random Tree Search

Mobile robots operating in dynamic environments require computationally efficient solutions to the trajectory planning problem that respect kinematic limits and minimize the navigation cost. In this work, we present an approach to plan local trajectories that incrementally advances the robot along a global path, taking into account its kinematic constraints and changes in the environment. The key idea of the presented approach is to combine space discretization and heuristic-based search into a single algorithm. In order to explore the local neighborhood, the algorithm generates a heuristic-driven trajectory tree by iteratively expanding the most promising vertex with randomly sampled control inputs. Then, the robot advances along the best local trajectory and the process is repeated from the new configuration. The approach was successfully applied in a real differential drive robot. Here, we demonstrate the efficacy of our method through tests performed in simulation. The results highlight our approach's suitability for mobile robot navigation in complex dynamic environments.