See but Don’t Be Seen: Towards Stealthy Active Search in Heterogeneous Multi-Robot Systems

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 an unknown number of sparsely located targets in an unknown environment. This formulation addresses the requirement that search missions should focus on quick recovery of targets rather than full coverage of the search region. Previous approaches fail to accurately model sensing uncertainty, account for occlusions due to foliage or terrain when estimating the next optimal action, meet the requirement for heterogeneous search teams or 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.

Some search problems have an additional dimension to seek targets while attempting to conceal the search agents' location from the targets. This is applicable to reconnaissance missions wherein the safety of the search agents can be compromised. Prior work usually focuses on adversarial search settings where the evaders (targets) are actively trying to evade the pursuers (search agents), however, most approaches assume unrealistic parameters such as complete knowledge, infinite travel speed, unlimited compute, and/or perfect observation models. We model the problem as a multi-objective optimisation over the potential information gain of taking an action and the risk of information leakage to an unknown number of targets with unknown locations. We present the Stealthy Topography-Aware Reconnaissance (STAR) algorithm, a multi-objective parallelized Thompson sampling-based algorithm that relies on a strong topographical prior to reason over changing visibility risk over the course of the search.

In both sub-problems defined above, we show through simulation experiments that GUTS and STAR consistently outperform existing methods in their respective problems. We conduct field tests using our multi-robot system in an unstructured environment with a search area of varying scale 0.075 sq. km - 2.6 sq. km). Our system demonstrates robustness to various failure modes, achieving full recovery of targets (where feasible) in every field test.