Popular navigation stacks implemented on top of open-source frameworks such as ROS(Robot Operating System) and ROS2 represent the robot workspace using a discretized 2D occupancy grid. This method, while requiring less computation, restricts the use of such navigation stacks to wheeled robots navigating on flat surfaces. In this paper, we present a navigation stack that uses a volumetric representation of the robot workspace, and hence can be extended to aerial and legged robots navigating through uneven terrain. Additionally, we present a new sampling-based motion planning algorithm which introduces a bi-directional approach to the Batch Informed Trees (BIT*) motion planning algorithm, whilst wrapping it with a strategy switching approach in order to reduce the initial time taken to find a path, in addition to the time taken to find the shortest path.
View on arXiv