default Towards fully autonomous underwater vehicles in ASW scenarios: an adaptive AUV mission management layer


Towards fully autonomous underwater vehicles in ASW scenarios: an adaptive AUV mission management layer.  Ferri, Gabriele. CMRE-FR-2015-006. June 2015.

In this work we investigate how to improve the decision making of AUVs for their effective use in a multistatic ASW network. In multistatic ASW an acoustic source insonifies a target (a submarine) and the reflected pulse is detected by a hydrophone line array towed by the receiving AUVs. The capability for an AUV to make autonomous decisions is crucial in this scenario, especially if we consider the limited bandwidth of underwater acoustic communications that makes communication with the vehicles limited and sometimes impossible. To be really effective, AUVs need to take decisions autonomously on the basis of the acquired data and the changing tactical scene. Data-driven approaches can increase the performance of the mission by allowing the AUV, for instance, to adapt its path to achieve some mission objectives. Recently we have designed and successfully tested at sea a non-myopic, receding horizon algorithm which controls the heading of the AUV to minimize the expected target position estimation error of a tracking filter. Minimizing this error is typically of the utmost interest in target state estimation since it is one way of maintaining track. A candidate track is used by the non-myopic algorithm to control the AUV to achieve favourable target-source-receiver geometries. The AUV has therefore to select tracks likely being target-generated. ASW scenarios are typically complex from the detection/tracking point of view. The target may not be observable for long time due to the particular sound speed profile or low probability of detection. Several false tracks are usually simultaneously present and may last for several pings and finally the presence of ambiguous tracks (due to the port-starboard ambiguity of contacts in line arrays) increases the number of tracks of possible interest. Only the most interesting tracks should be investigated without wasting time and energy to optimize tracks not target related. In this work we present an adaptive, data driven Mission Management Layer (MML) running on board the vehicles managing all the phases of the AUV missions. The MML receives the tracks and contacts produced by the signal processing chain, takes decisions in real-time on which tracks are interesting to be prosecuted and commands the vehicle control layer operations. First of all, a metric is needed to quantify the quality of a track. The track quality can be defined as the probability of existence of the target corresponding to the track. In this work we propose a track scoring method based on the quality of the measurement to- track associations. The method uses a model of the acoustics and the kinematic features of the target and does not need the knowledge of parameters that are difficult to estimate such as the probability of detection. The real-time track score can then be used to classify the tracks and select which ones are to be prosecuted by the non-myopic optimizer. The MML manages all the phases of an ASW mission: exploration of the area, disambiguation between a track and the relative ambiguous (ghost) track when one firmed track is present, optimization of a confirmed track and target reacquisition when a track breaks. A compromise is found between the exploration/surveillance of the area and behaviours that improve the tracking and classification performance on identified tracks. Only the most interesting tracks are prosecuted to avoid wasting time/energy in pursuing tracks not target generated. These features are necessary for effective data-driven behaviours in real ASW scenarios. Our mission management approach pushes towards the full autonomy of our system since it provides the AUV the capability of adapting its actions to the current tactical situation. In the work we start by proposing a taxonomy for an ASW mission from the AUV perspective. After the description of the track scoring and of the MML we describe the implementation of the proposed architecture in the MOOS-based control architecture of CMRE OEXs. We present results from sea trials (REP14 Atlantic and COLLAB-NGAS14) demonstrating the effectiveness of our approach. These results represent one of the first examples of AUVs autonomously taking decisions in a realistic, complex littoral surveillance scenario.