An agent to optimally re-distribute control in an underactuated AUV

Abstract

A knowledge-based agent was designed and validated to optimally re-distribute control authority in a torpedo-shaped autonomous underwater vehicle (AUV). The objective is greater fault tolerance in AUVs on long deployments when an AUV is unexpectedly underactuated from a jammed control fin. The optimisation is achieved through a genetic algorithm (GA) that evaluates solutions based on a full non-linear analysis of the AUV dynamics and control. The AUV dynamics, hydrodynamics, and control have to be well known ahead of time. The agent is implemented on-board the AUV to provide timely re-assignment of the fin control authority (gains), underway, and consequently the mission can continue or a potential vehicle loss averted. The effectiveness of the agent is assessed through a parametric analysis that compares the response of the unexpectedly underactuated AUV with its initial gains against the optimised gains. The agent’s greatest impact is in the event of a bow fin jam as the remaining three planes cannot depth-keep well without the agent.