Real-Time Distributed Infrastructure-free Searching and Target Tracking via Virtual Pheromones

TL;DR

This work tackles real-time distributed active perception in GPS-denied, infrastructure-free environments where agents operate with limited local information. It proposes an integrated algorithm that combines a decentralized virtual pheromone-based coverage control with a distributed greedy target selection to enable online searching and multi-target tracking. The approach propagates and fuses target state estimates via error-ellipse covariance propagation and maintains a decaying pheromone map to drive exploration, all within an $r_c$-disk communication framework and without global localization. Validation includes Monte-Carlo simulations and real hardware experiments with LTA robots, showing improved target discovery and tracking performance over baseline strategies while maintaining scalable computational demands on the order of $O(N_s^2 + N_h + w_i^{\text{init}} / w_i^\delta)$. These results demonstrate a practical, scalable solution for infrastructure-free, cooperative perception in challenging environments.

Abstract

Actively searching for targets using a multi-agent system in an unknown environment poses a two-pronged problem, where on the one hand we need agents to cover as much of the environment as possible with little overlap and on the other hand the agents must coordinate among themselves to select and track targets thereby maximizing detection performance. This paper proposes a fully distributed solution for an ad hoc network of agents to cooperatively search for targets and monitor them in an unknown infrastructure-free environment. The solution combines a distributed pheromone-based coverage control strategy with a distributed target selection mechanism. We further expand the scope to show the implementation of the proposed algorithm on a Lighter Than Air (LTA) multi-robotic system that can search and track objects in priori unknown locations.

Figures (17)

• Figure 1: Flow-chart of one iteration of solution running in agent.
• Figure 2: Target selection strategy. In this illustration, we show how the targets will be assigned when 2 or more targets are in the FOV (Agents 1 and 2) and how a target can be assigned in a neighboring agent's FOV (Agent 3).
• Figure 3: Pheromone Map. Orange shows the pheromones in the local pheromone list. Green points show the locations of pheromone in the neighbor pheromone list. The different shades of blue shows the different strengths of pheromone.
• Figure 4: Final target estimate is combined estimate of the data available in each agent. Here agent 1 combines its own estimate with that received from agent 2 using error ellipsis method
• Figure 5: Simulation of the active perception algorithm. The environment map is shown in (a) with agents in blue, FOVs in purple, targets in red, waypoints being the red lines and agent trajectories being the gray lines. In (b), we plot the $\|x_{i, k}(t) - \widehat{x}_{i, k}^c(t)\|$ and $\mathbf{h}(x_{i, k}|\mathcal{I}(0:t))$ as time passes.

Theorems & Definitions (2)

• Remark 3.1: Collision avoidance
• Remark 5.1: Sim. - Expm. Compatibility