Contingency planning for long-duration AUV missions

In recent years, the use of autonomous underwater vehicles has become increasingly popular for a wide variety of applications. As the cost of deploying a vehicle and the risk of loss or damage are often high, AUV missions typically consist of simple pre-scripted behaviours. Designed to minimise risk to the vehicle and its scientific cargo, these behaviours are inevitably overly-conservative, reserving a significant proportion of battery as a contingency should usage be higher than expected. Consequently, in the average case, the vehicle is not used to its full potential. As environments in which AUVs operate are dynamic and their effect on the vehicle is often uncertain, it is difficult to accurately predict the resource cost of a mission or individual task in advance. By modelling this uncertainty and allowing the vehicle to observe both the progress of the mission and the surrounding environment, the mission plan may be autonomously refined during operation. For example, in the event that resource usage, such as battery power, is observed to be lower than expected, the vehicle can schedule additional data collection tasks. Conversely, if the resource usage is higher than expected, the vehicle can remove lower priority tasks from the mission plan in order to increase the probability of successful recovery without the need to abort the mission. Such planning becomes increasingly beneficial when performing longer duration missions comprised of many tasks. This paper discusses the development of a new autonomous planning algorithm which models the uncertainty in the AUV domain and attempts to maximise the collection of scientific data without compromising the safety of the vehicle. It includes a technical overview, recent results and a discussion of the research in the context of potential applications, focusing on long-range and low-cost vehicles.