AquaHet-PSO: An Informative Path Planner for a Fleet of Autonomous Surface Vehicles with Heterogeneous Sensing Capabilities based on Multi-Objective PSO

Abstract

The importance of monitoring and evaluating the quality of water resources has significantly grown over time. To achieve this effectively, an option is to employ an intelligent monitoring system capable of measuring the physical and chemical parameters of water. Surface vehicles equipped with sensors for measuring water quality parameters offer a viable solution for these missions. This work presents a novel approach called AquaHet-PSO, which addresses the challenge of simultaneously monitoring multiple water quality parameters with several peaks of contamination using a heterogeneous fleet of autonomous surface vehicles. Each vehicle in the fleet possesses a different set of sensors, such as number of sensors and sensor types, which is the definition provided by the authors for a heterogeneous fleet. The AquaHet-PSO consists of three main phases. In the initial phase, the vehicles traverse the water resource to obtain preliminary models of water quality parameters. These models are then utilized in the second phase to identify potential contamination areas and assign vehicles to specific action zones. In the final phase, the vehicles focus on a comprehensive characterization of the parameters. The proposed system combines several techniques, including Particle Swarm Optimization and Gaussian Processes, with the integration of genetic algorithm to maximize the distances between the initial positions of vehicles equipped with identical sensors, and a distributed communication system in the final phase of the AquaHet-PSO. Simulation results in the Ypacarai lake demonstrate the effectiveness and efficiency of AquaHet-PSO in generating accurate water quality models and detecting contamination peaks. The proposed method demonstrated improvements compared to the lawnmower approach. It achieved a remarkable 17% improvement, on r-squared data, in generating complete models of water quality parameters throughout the lake. In addition, it achieved a 230% improvement in accurate characterization of high pollution areas and a 24% increase in pollution peak detection specifically for heterogeneous fleets equipped with four or more identical sensors.