Multi-Agent Robotic Exploration of Confined Aquatic Environments

Abstract

Exploratory missions in confined aquatic spaces often face significant challenges, including constraints in self-localisation, communication, and maneuverability, challenges that many current platforms fail to overcome. This thesis aims to address these issues by utilising a Cooperative Aquatic Vehicle Exploration System (CAVES). Firstly, an Image-Based Visual Servoing (IBVS) leader-follower control architecture for heterogeneous aquatic robots was proposed, which does not rely on underwater localisation and communications. Experiments were conducted in two different uncluttered indoor ponds. Results show that robots can maintain tracking of each other with maximum x and y displacements of 0.42 m and 0.41 m and with a constant z displacement of 1.2 m. Loss of Line-of-Sight (LoS) was also observed during the experiments, due to the intermittent measurements with an underwater camera, which can severely impact the robustness of the system. Secondly, to improve the robustness of traditional multi-agent collaborative control approaches that utilise unreliable measurements, the concept of a Virtual Elastic Tether (VET) was proposed. VET simulates an imaginary tether to connect multiple robots using mutual camera views, allowing them to execute their individual maneuvers while maintaining a defined level of mutual proximity. Experiments were conducted in both simulation and in an uncluttered pond, benchmarked against a traditional IBVS approach. Results indicate that while the leader-follower formation of the baseline system failed under external perturbations, the VET-enhanced system's formation recovered to its pre-perturbation state within 5 seconds. Moreover, the VET-enhanced CAVES successfully navigated a confined water pond where the baseline approach failed to perform adequately. However, these experiments were conducted in uncluttered ponds, whereas practical applications may involve ponds containing unknown objects. For safe navigation in cluttered environments, a novel dog walking paradigm was proposed. This paradigm models the implicit communication-based interaction between a dog walker (the leader) and the dog (the follower). The behavioral patterns between aquatic robots were then modeled under such paradigm. Simulations were conducted, and performance was benchmarked against a setup using only VET. Results indicate that the dog walking paradigm significantly enhances cooperative behavior among agents. The outcomes of this thesis suggest that the proposed approaches could potentially be applied to underwater exploratory missions. However, the formulations of the VET and the dog walking paradigm are not fully optimised and rely on a set of tunable parameters, which may limit their effectiveness and necessitate further refinement and modeling in future research.

Date of Award	6 Jan 2025
Original language	English
Awarding Institution	The University of Manchester
Supervisor	Simon Watson (Supervisor) & Ognjen Marjanovic (Supervisor)