A Tree-based Next-best-trajectory Method for 3D UAV Exploration

TL;DR

The paper tackles autonomous 3D UAV exploration in completely unknown environments by integrating exploration and path planning into a single, tree-based framework called Exploration-RRT (ERRT). ERRT generates candidate goals, expands a robot-safe RRT* locally, computes model-based actuation trajectories via NMPC, and evaluates trajectories using information gain along the entire path to select the next move. Key contributions include a novel momentary Next-Best-Trajectory formulation, 3D safe path generation with UAV dynamics, information-gain evaluation along trajectories, UFOmap integration for unknown space, and open-source software with extensive simulations and field tests against state-of-the-art planners. The approach demonstrates real-time performance on constrained hardware and shows improved exploration efficiency in both subterranean and urban environments, offering practical impact for GPS-denied, cluttered settings.

Abstract

This work presents a fully integrated tree-based combined exploration-planning algorithm: Exploration-RRT (ERRT). The algorithm is focused on providing real-time solutions for local exploration in a fully unknown and unstructured environment while directly incorporating exploratory behavior, robot-safe path planning, and robot actuation into the central problem. ERRT provides a complete sampling and tree-based solution for evaluating "where to go next" by considering a trade-off between maximizing information gain, and minimizing the distances travelled and the robot actuation along the path. The complete scheme is evaluated in extensive simulations, comparisons, as well as real-world field experiments in constrained and narrow subterranean and GPS-denied environments. The framework is fully ROS-integrated, straight-forward to use, and we open-source it at https://github.com/LTU-RAI/ExplorationRRT.

Figures (25)

• Figure 1: Local Exploration over 6 minutes in the DARPA Final Stage gazebo world - mimicking the conditions of the real competition. Figures highlighting the progress of exploration at set times. Total exploration path length was around $\unit[250]{m}$.
• Figure 2: Two more ERRT exploration runs into different parts of the area, with varying kinds of environments from urban warehouse like areas to tunnels and caves.
• Figure 3: The ERRT process - 1) local robot-safe Tree Expansion filling $V^l_\mathrm{safe}$, 2) pseudo-random goal sampling $g^\mathrm{c}$ (yellow dots) and improved actuation-paths $\bm{\chi}_j$ (green), and 3) selected path $\bm{\chi}^*$(green) with marked unknown voxels that will be discovered along the "next-best-trajectory" (red) (bottom).
• Figure 4: Robot-safe 3D Tree expansion in an obstacle-rich and complex warehouse-like area (top), and selected exploration "next-best-trajectory" (green line) (bottom). The "clumps" of yellow dots are the candidate goals and define the areas of interest to visit as they have information gain.
• Figure 5: Explored area after 15 minutes in the DARPA Cave World - approx. 10-15m wide tunnels. 3D view highlighting the vertical structures (top) and the map overview (bottom).