Remote submersibles

Controlling the cooperative behaviors of a fleet of autonomous underwater vehicles in a stochastic, complex environment is a formidable challenge in artificial intelligence. The complexity arises from the challenges of limited navigation and communication capabilities of underwater environment. A time critical cooperative operation by acoustic networks of Multiple Cooperative Vehicles (MCVs) necessitates a robust task allocation mechanism and an efficient path planning model. In this work, we present solutions to investigate two aspects of the cooperative schema for multiple underwater vehicles under realistic underwater acoustic communications: a Location-aided Task Allocation Framework (LAAF) algorithm for multi-target task assignment and a mathematical programming model, the Grid-based Multi-Objective Optimal Programming (GMOOP), for finding an optimal vehicle command decision given a set of objectives and constraints. We demonstrate that, the location-aided auction strategies perform significantly better than the generic auction algorithm in terms of effective task allocation time and information bandwidth requirements. In a typical task assignment scenario, the time needed in the LAAF algorithm is only a fraction compared to the generic auction algorithm. On the other hand; the GMOOP path planning technique provides a unique means for multi-objective tasks by cooperative agents with limited communication capabilities. Under different environmental settings, the GMOOP path planning technique is proved to provide a method with balance of sufficient expressive power and flexibility, and its solution algorithms tractable in terms of mission completion time, with a limited increase of overhead in acoustic communication. Prior to this work, existing multi-objective action selection methods were limited to robust networks where constant communication available.

Acoustic networks of autonomous underwater vehicles (AUVs) show great promise, but a lack of simulation tools and reliance on protocols originally developed for terrestrial radio networks has hindered progress. This work addresses both issues. A new simulator of underwater communication among AUVs provides accurate communication modeling and flexible vehicle behavior, while a new routing protocol, location-aware source routing (LASR) provides superior network performance. The new simulator was used to evaluate communication without networking, and then with networking using the coding or dynamic source routing (DSR) protocols. The results confirmed that a network was essential to ensure effective fleet-wide communication. The flooding protocol provided extremely reliable communication but with low message volumes. The DSR protocol, a popular routing protocol due to its effectiveness in terrestrial radio networks, proved to be a bad choice in an acoustic environment: in most cases, it suffered from both poor reliability and low message volumes. Due to the high acoustic latency, even moderate vehicle speeds caused the network topology to change faster than DSR could adapt. DSR's reliance on shortest-path routing also proved to be a significant disadvantage. Several DSR optimizations were also tested; most proved to be unhelpful or actually harmful in an underwater acoustic network. LASR was developed to address the problems noted in flooding and DSR. LASR was loosely derived from DSR, most significantly retaining source routes and the reply/request route discovery technique. However, LASR added features which proved, in simulation, to be significant advantages -- two of the most effective were a link/route metric and a node tracking system. To replace shortest-path routing, LASR used the expected transmission count (ETX) metric.