GUTS: Generalized Uncertainty-Aware Thompson Sampling for Multi-Agent Active Search

TL;DR

The Generalized Uncertainty-aware Thompson Sampling (GUTS) algorithm is presented, which is suitable for deployment on heterogeneous multi-robot systems for active search in large unstructured environments and demonstrates robustness to various failure modes.

Abstract

Robotic solutions for quick disaster response are essential to ensure minimal loss of life, especially when the search area is too dangerous or too vast for human rescuers. We model this problem as an asynchronous multi-agent active-search task where each robot aims to efficiently seek objects of interest (OOIs) in an unknown environment. This formulation addresses the requirement that search missions should focus on quick recovery of OOIs rather than full coverage of the search region. Previous approaches fail to accurately model sensing uncertainty, account for occlusions due to foliage or terrain, or consider the requirement for heterogeneous search teams and robustness to hardware and communication failures. We present the Generalized Uncertainty-aware Thompson Sampling (GUTS) algorithm, which addresses these issues and is suitable for deployment on heterogeneous multi-robot systems for active search in large unstructured environments. We show through simulation experiments that GUTS consistently outperforms existing methods such as parallelized Thompson Sampling and exhaustive search, recovering all OOIs in 80% of all runs. In contrast, existing approaches recover all OOIs in less than 40% of all runs. We conduct field tests using our multi-robot system in an unstructured environment with a search area of approximately 75,000 sq. m. Our system demonstrates robustness to various failure modes, achieving full recovery of OOIs (where feasible) in every field run, and significantly outperforming our baseline.

Figures (5)

• Figure 1: Test-field in Pittsburgh, PA. (a) shows the traversal costmap and the search polygon (purple is high cost and black is low cost), and (b) shows an overhead image with the most common OOI locations and launching areas of the robots. The area inside the polygon is roughly 75,000 sq meters.
• Figure 2: Left: RecBot UGV. Right: Lil Hexy UAV.
• Figure 3: Left: Top-down view of UGV camera field-of-view. Span is 192° horizontally and 37° vertically. Right: UAV camera field-of-view. Span is $19\degree \times 16 \degree$. At a flight height of 80m, this leads to approximately $27.8m \times 22.5m$ rectangle of visibility on the ground.
• Figure 4: Object detector: qualitative results. We show images and OOI detections for the UGV (left) and the UAV (right)
• Figure 5: Comparison of Search Efficiency. (Left) We visualize the search efficiency for each run in Table I. We plot the recall versus the number of sensing actions taken per robot. This graph shows that GUTS outperforms the coverage baseline on our system. We also observe that subsampling the total set of waypoints achieves a good balance of computational efficiency and performance, and slighly outperforms the coverage baselines on UAVs. (Middle and Right) We compare the recall rate of different search algorithms vs runtime in a single-robot and multi-robot simulation. We again observe that GUTS outperforms NATS and coverage-based search.