Autonomous 3D Exploration in Large-Scale Environments with Dynamic Obstacles

TL;DR

DAEP addresses autonomous 3D exploration in dynamic environments by extending AEP with a Kalman-predictor-based trajectory model and time-aware planning. It introduces a dynamic score that fuses dynamic information gain, travel cost, and a Dynamic Frequency Map, and uses time-based RRTs to foresee obstacle trajectories. A new dynamic benchmark with ten large-scale scenarios plus extensive evaluations demonstrates that DAEP outperforms static and dynamic baselines in exploration efficiency and collision avoidance, including in the Village outdoor environment. The work advances practical capabilities for safe, scalable exploration in real-world, time-varying settings and suggests future integration with advanced motion planning for field deployments.

Abstract

Exploration in dynamic and uncertain real-world environments is an open problem in robotics and constitutes a foundational capability of autonomous systems operating in most of the real world. While 3D exploration planning has been extensively studied, the environments are assumed static or only reactive collision avoidance is carried out. We propose a novel approach to not only avoid dynamic obstacles but also include them in the plan itself, to exploit the dynamic environment in the agent's favor. The proposed planner, Dynamic Autonomous Exploration Planner (DAEP), extends AEP to explicitly plan with respect to dynamic obstacles. To thoroughly evaluate exploration planners in such settings we propose a new enhanced benchmark suite with several dynamic environments, including large-scale outdoor environments. DAEP outperform state-of-the-art planners in dynamic and large-scale environments. DAEP is shown to be more effective at both exploration and collision avoidance.

Figures (6)

• Figure 1: Real world 3D environments can be non-trivial to explore in detail due to large scales, geometrical complexity and the presence of dynamic obstacles wandering about.
• Figure 2: System overview for DAEP. It consists of a local and global planner that collaborate to explore the dynamic environment in an efficient and safe way.
• Figure 3: Dynamic information gain. The white circle represents the current position of a dynamic obstacle while the grey square represents its position five seconds into the future. The blue rays represents the field of view of the agent as well as the information gain that can be acquired from that pose. The red rays represents the excluded information gain since it will not be visible upon arrival to the node (top) due to the dynamic obstacle blocking the view.
• Figure 4: View of the Village area. The red bounding box illustrates the volume to be explored. The area is roughly 1 hectare and populated with multiple dynamic obstacles. The paths of the dynamic obstacles are highlighted in green where the lines indicate that the dynamic obstacle move back and forth while a loop indicates that the dynamic obstacles circulate.
• Figure 5: Upper row: Exploration progress in Maze. The maximum volume is 865 $m^3$. Here AEP halts the exploration where the envelope stops. Lower row: Exploration progress in the Village environment to the left and collision graphs in the middle and to the right for Maze. The collision graphs presents the run that accumulated the most coverage. Orange segments indicate collisions and their duration.