Reactive evolutionary path planning for autonomous surface vehicles in lake environments

Resumen

Autonomous Surface Vehicles (ASVs) have found a lot of promising applications in aquatic environments, i.e., sea, lakes, rivers, etc. They can be used for applications of paramount importance, such as environmental monitoring of water resources, and for bathymetry to study the characteristics of the basing of a lake/sea or for surveillance in patrol missions, among others. These vehicles can be built with smaller dimensions when compared to regular ships since they do not need an on-board crew for operation. However, they do require at least a telemetry control as well as certain intelligence for making decisions and responding to changing scenarios. Water resources are very important in Paraguay since they provide fresh water for its inhabitants and they are crucial for the main economic activities such as agriculture and cattle raising. Furthermore, they are natural borders with the surrounding countries, and consequently the main transportation route for importing/exporting products. In fact, Paraguay is the third country in the world with the largest fleet of barges after USA and China. Thus, maintaining and monitoring the environmental conditions of these resources is key in the development of the country. This work is focused on the maintenance and monitoring of the greatest lake of the country called Ypacarai Lake. In recent years, the quality of its water has been seriously degraded due to the pollution caused by the low control of the dumping of waste thrown into the Lake. Since it is also a national icon, the government of Paraguay has put a lot of effort in recovering water quality of the Lake. As a result, it is monitored periodically but using manual procedures. Therefore, the primary objective of this work is to develop these monitoring tasks autonomously by means of an ASV with a suitable path planning strategy. Path planning is an active research area in robotics. A particular case is the Coverage Path Planning (CPP) problem, where an algorithm should find a path that achieves the best coverage of the target region to be monitored. This work mainly studies the global CPP, which returns a suitable path considering the initial conditions of the environment where the vehicle moves. The first contribution of this thesis is the modeling of the CPP using Hamiltonian Circuits (HCs) and Eulerian Circuits (ECs). Therefore, a graph adapted to the Ypacarai Lake is created by using a network of wireless beacons located at the shore of the lake, so that they can be used as data exchange points between a control center and the ASV, and also as waypoints. Regarding the proposed modeling, HCs and ECs are paths that begin and end at the same point. Therefore, the ASV travels across a given graph that is defined by a set of wireless beacons. The main difference between HC and EC is that a HC is a tour that visits each vertex only once while EC visits each edge only once. Finding optimal HCs or ECs that minimize the total distance traveled by the ASV are very complex problems known as NP-complete. To solve such problems, a meta-heuristic algorithm can be a suitable approach since they provide quasi-optimal solutions in a reasonable time. In this work, a GA (Genetic Algorithm) approach is proposed and tested. First, an evaluation of the performance of the algorithm with different values of its hyper-parameters has been carried out. Second, the proposed approach has been compared to other approaches such as randomized and greedy algorithms. Third, a thorough comparison between the performance of HC and EC based approaches is presented. The simulation results show that EC-based approach outperforms the HC-based approach almost 2% which in terms of the Lake size is about 1.4 km2 or 140 ha (hectares). Therefore, it has been demonstrated that the modeling of the problem as an Eulerian graph provides better results. Furthermore, it has been investigated the relationship between the number of beacons to be visited and the distance traveled by the ASV in the EC-based approach. Findings indicate that there is a quasi-lineal relationship between the number of beacons and the distance traveled. The second contribution of this work is the development of an on-line learning strategy using the same model but considering dynamic contamination events in the Lake. Dynamic events mean the appearance and evolution of an algae bloom, which is a strong indicator of the degradation of the lake. The strategy is divided into two-phases, the initial exploration phase to discover the presence of the algae bloom and next the intensification phase to focus on the region where the contamination event is detected. This intensification effect is achieved by modifying the beacon-based graph, reducing the number of vertices and selecting those that are closer to the region of interest. The simulation results reveal that the proposed strategy detects two events and monitors them, keeping a high level of coverage while minimizing the distance traveled by the ASV. The proposed scheme is a reactive path planning that adapts to the environmental conditions. This scheme makes decisions in an autonomous way and it switches from the exploratory phase to the intensification phase depending on the external conditions, leading to a variable granularity in the monitoring task. Therefore, there is a balance between coverage and the energy consumed by the ASV. The main benefits obtained from the second contribution includes a better monitoring in the quality of water and control of waste dumping, and the possibility to predict the appearance of algae-bloom from the collected environmental data.