Multi-Robot Target Tracking with Sensing and Communication Danger Zones

TL;DR

The paper tackles robust multi-robot active target tracking in environments with sensing and communication attacks by formulating a centralized problem subject to probabilistic safety constraints. It introduces two danger-zone models (sensing and communication) with Gaussian-uncertainty sources and converts the resulting chance constraints into deterministic bounds using linearization and $\operatorname{erf}^{-1}$-based terms, enabling online planning. An EKF-based target estimation component is integrated into a nonlinear objective that minimizes target-state uncertainty while penalizing control effort, solved via FORCESNLP. The approach is validated through MATLAB simulations and hardware experiments with aerial and ground robots, demonstrating risk-aware behavior and practical resilience, though sim-to-real transfer challenges remain.

Abstract

Multi-robot target tracking finds extensive applications in different scenarios, such as environmental surveillance and wildfire management, which require the robustness of the practical deployment of multi-robot systems in uncertain and dangerous environments. Traditional approaches often focus on the performance of tracking accuracy with no modeling and assumption of the environments, neglecting potential environmental hazards which result in system failures in real-world deployments. To address this challenge, we investigate multi-robot target tracking in the adversarial environment considering sensing and communication attacks with uncertainty. We design specific strategies to avoid different danger zones and proposed a multi-agent tracking framework under the perilous environment. We approximate the probabilistic constraints and formulate practical optimization strategies to address computational challenges efficiently. We evaluate the performance of our proposed methods in simulations to demonstrate the ability of robots to adjust their risk-aware behaviors under different levels of environmental uncertainty and risk confidence. The proposed method is further validated via real-world robot experiments where a team of drones successfully track dynamic ground robots while being risk-aware of the sensing and/or communication danger zones.

Figures (5)

• Figure 1: (a) Geometrical illustration for a communication danger zone where the position of the central attacking source is non-deterministic. The uncertainty is reflected as a blue radial gradient; (b) Shows the original integration region corresponding to the chance constraints; (c) Linearizes the integration region to compute an upper-bound of the multivariate integral. The line $\mathbf{a}^\top \mathbf{x} = b$ is tangent to the circle.
• Figure 2: Risk-aware tracking with a sensing danger zone. Each row corresponds to one selection of the combination ($\mathbf{\Sigma}$, $\epsilon_1$), and three subfigures in the same row show the corresponding tracking process. Green lines are the trajectories of robots and black lines are the trajectories of targets. We use light and dark green for the two robots respectively. The initial state of robots and targets are drawn as dots with the corresponding colors. The red area represents a Gaussian distribution with an associated mean position in dark red dots. A more spread-out distribution corresponds to a larger $\mathbf{\Sigma}$ in the second row. The dotted circle represents the safety clearance as the radius.
• Figure 3: (a)(b): trace and probability of sensor failure throughout the tracking process with a Sensing danger zone; (c)(d) trace and probability of communication jamming with a Communication danger zone. The comparison of results under three sets of parameters is shown.
• Figure 4: Risk-aware tracking with a communication danger zone. Each row corresponds to a tracking case under one parameter setting, and three sub-figures in the same row show the tracking process under that setting. Robots' trajectories are plotted in dark and light green, and the targets' trajectories are plotted in black. The two targets move in circles in a counter-clockwise direction. Robots' initial positions are shown as green dots. The first column also shows the initial positions of targets as black dots. The blue area represents that the position of the jamming source follows a Gaussian distribution. Uncertainty of the jamming source's position is reflected by how spread-out the area is, i.e., a larger $\mathbf{\Sigma}$ leads to a more spread-out area. Dark blue dots denote the mean position of the jamming source.
• Figure 5: Screenshots of hardware experiments. (a) 2 robots track 2 targets with a sensing danger zone; (b) 4 robots track 2 targets with a communication danger zone; (c) 3 robots track 3 targets with both types of danger zones. The video of the experiments is available at: https://youtu.be/uSYPI817Y6c