Decentralized Multi-Agent Active Search and Tracking when Targets Outnumber Agents

TL;DR

This work tackles decentralized, asynchronous multi-agent active search and tracking when targets outnumber agents and no central controller is available. It introduces DecSTER, which fuses a sequential Monte Carlo implementation of the PHD filter with Thompson sampling to drive explore-exploit decisions in a continuous search space. Empirical results show that TS-PHD-II, which samples target cardinality from a Poisson distribution and locations from a mixture, achieves superior $OSPA$ performance across varying team sizes and target counts and remains robust to communication delays. The approach advances practical, scalable multi-agent search-and-track under uncertainty and lays groundwork for real-world deployments on autonomous systems.

Abstract

Multi-agent multi-target tracking has a wide range of applications, including wildlife patrolling, security surveillance or environment monitoring. Such algorithms often make restrictive assumptions: the number of targets and/or their initial locations may be assumed known, or agents may be pre-assigned to monitor disjoint partitions of the environment, reducing the burden of exploration. This also limits applicability when there are fewer agents than targets, since agents are unable to continuously follow the targets in their fields of view. Multi-agent tracking algorithms additionally assume inter-agent synchronization of observations, or the presence of a central controller to coordinate joint actions. Instead, we focus on the setting of decentralized multi-agent, multi-target, simultaneous active search-and-tracking with asynchronous inter-agent communication. Our proposed algorithm DecSTER uses a sequential monte carlo implementation of the probability hypothesis density filter for posterior inference combined with Thompson sampling for decentralized multi-agent decision making. We compare different action selection policies, focusing on scenarios where targets outnumber agents. In simulation, we demonstrate that DecSTER is robust to unreliable inter-agent communication and outperforms information-greedy baselines in terms of the Optimal Sub-Pattern Assignment (OSPA) metric for different numbers of targets and varying teamsizes.

Figures (5)

• Figure 1: Problem setup. (a) Agents sense different regions of the search space at different vertical heights, receiving noisy 2D location coordinates of the possible targets in their field of view, along with false positive measurements. The targets shown as black crosses move in the search space with different velocities shown by the red arrows. (b) The line at the top indicates the target's continuous motion with time. In our asynchronous multi-agent setup, agents can collect observations without waiting for their teammates whereas in the synchronous setting, the solid boxes indicate the agents' idle wait times.
• Figure 2: Comparing the proposed TS methods. $\mathrm{DecSTER}$-II using TS-PHD-II consistently outperforms $\mathrm{DecSTER}$-I using TS-PHD-I. Increasing team size $J$ reduces the number of measurements per agent required to achieve similar OSPA.
• Figure 3: Baseline comparisons. For different numbers of targets and with fewer agents than targets, $\mathrm{DecSTER}$ with TS-PHD-II (denoted $\mathrm{DecSTER}$-II) outperforms random sensing (RANDOM) and information greedy baselines (RENYI, TS-RENYI) by achieving a lower OSPA for the same number of measurements per agent.
• Figure 4: DecSTER vs. DecSTER-C. $\mathrm{DecSTER}$-C optimizes only for the cardinality error in the OSPA objective. $\mathrm{DecSTER}$ outperforms $\mathrm{DecSTER}$-C indicating the advantage of jointly optimizing detection and localization errors with TS-guided explore-exploit decisions.
• Figure 5: Robustness to unreliable communication. When agents communicate their actions and observations with decreasing probability $p$, $\mathrm{DecSTER}$ experiences a graceful deterioration in OSPA performance and agents require increasingly more measurements to estimate the number and locations of true targets in the search space.