NavPi: An Adaptive Local Path-Planning Pipeline for 3D Navigation in Difficult Terrain

Mobile robots have become increasingly popular in various areas of everyday life. In order to increase their capabilities, the robots have to be able to autonomously navigate through their environment. This process is often divided into two steps. First a global path planner calculates a optimal collision free path on a prerecorded map. Afterwards the local path planner is used to perform a collision free execution of the generated global path. Since the global path planning is performed on a pre-recorded and therefore possibly outdated map, the planner also has to handle various unforeseen circumstances like dynamic obstacles or structural changes of the environment. In this work we present a system overview of our adaptive local path planning approach for mobile ground robots, that can handle these challenges. The core part of the presented approach is the plugin based pipeline structure, in which the individual components can be exchanged during runtime to handle changing environments and scenarios. Furthermore we present an three-dimensional obstacle avoidance strategy based on volumetric VDB-Maps, which enables the robot to traverse even the most challenging terrain. The local path planning system presented in this paper has been successfully applied to multiple robot systems in various real-world scenarios. The system has been tested on a range of use cases, from simple robots in storage facilities to walking robots traversing deserts and lunar-like environments during multiple field tests. The real-world application of the approach has demonstrated its ability to efficiently traverse challenging terrain while simultaneously avoiding collisions with unforeseen obstacles.