DWA-3D: A Reactive Planner for Robust and Efficient Autonomous UAV Navigation in Confined Environments

TL;DR

The paper addresses robust autonomous UAV navigation in cluttered, confined environments by introducing DWA-3D, a 3D extension of the Dynamic Window Approach that serves as a reactive local planner integrated with an RRT*-based global planner. It relies on online Octomap occupancy mapping from a 3D LiDAR and is validated through extensive real-world experiments, demonstrating bounded computation times and effective 3D obstacle avoidance in diverse scenarios. Key contributions include a velocity-space optimization with a nonuniform obstacle-distance measure and a theory-guided parameter tuning method that yields stable performance across environments. The work shows substantial practical impact for autonomous operation in tunnels, warehouses, galleries, and similar confined spaces, with a ROS-based framework and plans to release code for reproducibility and further development.

Abstract

Despite the growing impact of Unmanned Aerial Vehicles (UAVs) across various industries, most of current available solutions lack for a robust autonomous navigation system to deal with the appearance of obstacles safely. This work presents an approach to perform autonomous UAV planning and navigation in scenarios in which a safe and high maneuverability is required, due to the cluttered environment and the narrow rooms to move. The system combines an RRT* global planner with a newly proposed reactive planner, DWA-3D, which is the extension of the well known DWA method for 2D robots. We provide a theoretical-empirical method for adjusting the parameters of the objective function to optimize, easing the classical difficulty for tuning them. An onboard LiDAR provides a 3D point cloud, which is projected on an Octomap in which the planning and navigation decisions are made. There is not a prior map; the system builds and updates the map online, from the current and the past LiDAR information included in the Octomap. Extensive real-world experiments were conducted to validate the system and to obtain a fine tuning of the involved parameters. These experiments allowed us to provide a set of values that ensure safe operation across all the tested scenarios. Just by weighting two parameters, it is possible to prioritize either horizontal path alignment or vertical (height) tracking, resulting in enhancing vertical or lateral avoidance, respectively. Additionally, our DWA-3D proposal is able to navigate successfully even in absence of a global planner or with one that does not consider the drone's size. Finally, the conducted experiments show that computation time with the proposed parameters is not only bounded but also remains stable around 40 ms, regardless of the scenario complexity.

Figures (33)

• Figure 1: Our hexarotor performing real autonomous flights. In our facilities (top), accurate localization is provided by a motion capture system. Outside our laboratory (bottom), we perform SLAM with the onboard 3D LiDAR (FLOAM FLOAM).
• Figure 2: System General Scheme.
• Figure 3: Potential scenarios in which the drone could be deployed. The adverse visibility conditions would make a visual-based localization method fail. Despite the dense dust and lack of light, F-LOAM (FLOAM) was able to localize the UAV during this real flight.
• Figure 4: Attitude estimation drift situation in which a door gets blocked in the Octomap.
• Figure 5: Steps from the real world to the Octomap environment representation.