Localization uncertainty-aware autonomous exploration and mapping with aerial robots using receding horizon path-planning

This work presents an uncertainty-aware path-planning strategy to achieve autonomous aerial robotic exploration of unknown environments while ensuring mapping consistency on-the-go. The planner follows a paradigm of hierarchically optimized objectives, which are executed in receding horizon fashion. Initially, a random tree over the known feasible configurations is used to derive a maximal-exploration path, and its first viewpoint is selected as the next waypoint. Subsequently, an uncertainty-optimization step takes place, constructing within a local volume region a second tree of admissible alternative trajectories that all arrive at the reference viewpoint. Belief propagation of the robot state and the tracked landmarks in the environment takes place over the branches of this tree, and the path that minimizes the expected localization and mapping uncertainty is selected. This path is followed by the robot, and the entire process is iteratively repeated. The algorithm's computational complexity is analyzed and experimental results are used to evaluate its realtime execution efficiency onboard a micro aerial vehicle. The architecture of the complete pipeline is detailed, and an open-source implementation is provided. A complete aerial robot synthesis that enables high-fidelity autonomous reconstruction supported by the proposed planner is also elaborated. Comprehensive experimental evaluation studies that include mockup environments in ambient illumination and in challenging conditions such as clutter and darkness, as well as a field deployment in a railroad tunnel degraded visual environment are presented, with all data provided as openly available.