Aerial Robots Persistent Monitoring and Target Detection: Deployment and Assessment in the Field

TL;DR

This work tackles persistent monitoring and target detection with multiple UAVs in a bounded area under realistic disturbances. It proposes a distributed framework that fuses Time-inverted Kuramoto dynamics, 3D Lissajous curves, and Model Predictive Control to coordinate UAVs without precomputed trajectories. Key contributions include theoretical insights into $1$- and $\kappa$-cluster equilibria, the use of 3D Lissajous knots to enhance safety and coverage, a failure-resilient mechanism for both short and long disruptions, and field validation up to 11 UAVs. The approach offers a scalable, robust solution for real-time sensing tasks with strong practical relevance for drone swarms in dynamic environments.

Abstract

In this article, we present a distributed algorithm for multi-robot persistent monitoring and target detection. In particular, we propose a novel solution that effectively integrates the Time-inverted Kuramoto model, three-dimensional Lissajous curves, and Model Predictive Control. We focus on the implementation of this algorithm on aerial robots, addressing the practical challenges involved in deploying our approach under real-world conditions. Our method ensures an effective and robust solution that maintains operational efficiency even in the presence of what we define as type I and type II failures. Type I failures refer to short-time disruptions, such as tracking errors and communication delays, while type II failures account for long-time disruptions, including malicious attacks, severe communication failures, and battery depletion. Our approach guarantees persistent monitoring and target detection despite these challenges. Furthermore, we validate our method with extensive field experiments involving up to eleven aerial robots, demonstrating the effectiveness, resilience, and scalability of our solution.

Figures (18)

• Figure 1: Experimental verification of the proposed approach in the field with $11$ UAVs.
• Figure 2: Problem illustration. The rectangular space of dimension $[-A,A] \times [-B,B]$ indicates the area of interest. Cyan circles indicate the robots' positions, $\mathcal{L}$ indicates the path on which the robots are constrained to move, $\mathcal{R}$ describes the communication topology, and $r_s$ represents the sensing radius, equal for each robot.
• Figure 3: Minimum distance between robots in different configurations. In blue the results obtained by relying on borkar2016collision and 2D Lissajous curves. In orange the Time-inverted Kuramoto algorithm on 3D Lissajous curves. In green the Time-inverted Kuramoto algorithm on 2D Lissajous curves with incorrect initial configuration. Finally in red, borkar2016collision with incorrect initial configuration on 2D Lissajous curves.
• Figure 4: Overall block diagram for the $i$--th robot.
• Figure 5: Experiment with $7$ UAVs. The dashed lines represent the distances between two adjacent robots $d_{ij}$ in the x-y plane in time. The solid lines are the theoretical lower and upper bounds for safety and target detection guarantees, respectively. The dashed dot line is two times the sensing radius adjusted by a factor $\eta= 1.05$.

Theorems & Definitions (12)

• Remark 1
• Theorem 1: Convergence to the equilibrium boldrer2022multiagent
• Definition 1: Cluster boldrer2022multiagent
• Lemma 1: $\kappa$-clustered coverage boldrer2022multiagent
• Definition 2: Non-degenerate Lissajous curve
• Lemma 2: Lissajous knot
• proof
• Remark 2
• Remark 3: 3D Lissajous curve
• Theorem 2: Resiliency to type I failures boldrer2022time