Using Ambient Sensor Data to Characterize and Predict Autonomous Perception Sensor Performance

With the development of low-cost, small Uncrewed Underwater Vehicles (UUVs), the use of cost-efficient sensor systems imposes its own constraints (e.g., sensor accuracy and precision). Such sensors include Mechanically Scanning Imaging Sonars (MSIS), Conductivity-Temperature-Depth (CTD), and Commercial-off-the-Shelf (COTS) stereo camera systems, all of which are widely accessible perception sensors. To accommodate for the disadvantages that these low-cost sensors impose, the authors present a methodology to predict the accuracy and precision of specific perception sensor measurements in practical field implementation by leveraging the redundancy of several common on-board sensors. This insight can then be used to determine UUV region coverage control to allow for an ample collection of data samples (e.g., greater UUV loitering over areas with less-reliable data collection to ensure greater sample quantity and quality). Overall, the aim of this paper is to show how low-cost UUV sensor systems can be implemented without sacrificing sensor performance, particularly when compared to using their more expensive counterparts. This methodology developed by the authors is presented as the Standardized Heuristic Algorithm for Reinforced Calculations (SHARC). Here, the authors use an analytical model to show proof of concept of the validity of this proposed method. To test the SHARC algorithm, this study focuses on modeling a Mechanically Scanning Imaging Sonar (MSIS) in the BELLHOP simulation program and uses historic Sound Velocity Profiles (SVPs) to identify how the MSIS is affected by various ambient surroundings. Results show that the SVP shape affects the MSIS range of the probability of detection. It is observed that a change in SVP slope correlates to a reduced MSIS performance range as opposed to that of a more constant SVP depth profile, which increases the MSIS performance range. The authors show that SHARC enables the use of low-cost sensors to compete with higher capability counterparts by leveraging a priori data analysis to gauge in situ data performance and applies these techniques to a simulated MSIS to validate.