Towards real-time search planning in subsea environments

Abstract

We address the challenge of computing search paths in real-time for subsea applications where the goal is to locate an unknown number of targets on the seafloor. Our approach maximizes a formal definition of search effectiveness given finite search effort. We account for false positive measurements and variation in the performance of the search sensor due to geographic variation of the seafloor. We compare near-optimal search paths that can be computed in real-time with optimal search paths for which real-time computation is infeasible. We show how sonar data acquired for locating targets at a specific location can also be used to characterize the performance of the search sonar at that location. Our approach is illustrated with numerical experiments where search paths are planned using sonar data previously acquired from Boston Harbor.

Adaptive sampling with an autonomous underwater vehicle in static marine environments

“…These two planning algorithms are presented and compared because of their respective strengths and weaknesses in different applications. The foundations of branch‐and‐bound techniques are well‐suited for efficient informative path planning (Binney & Sukhatme, 2012; McMahon et al, 2017), particularly on low‐cost platforms with limited computational resources. However, the generation of the candidate planning tree is often dependent on suboptimal heuristics.…”

Acquiring environmental information yields better anticipated search performance

“…The environment in each cell is from a finite set of possible environments w 1 , w 2 , ... , w m . In practice, clustering of different types of environmental conditions can be carried out by using a previously acquired environment dataset in the search domain (see, for example, [30]). We presume that the actual environmental condition in each cell is not known, but that a probability distribution is known for each cell.…”

“…Since computingP lj \qi in (43) can be very expensive when the length of a line is large, we instead compute an empirical estimate ofP lj \qi as described in Section V-D. For each cell in the search grid, we performN trials to compute an empirical estimate ofP lj \qi , where in each trial we sample environment measurements from the remaining cells in line l j . After approximating the characterization gain for each cell individually, we apply an exact branch-and-bound method (similar to our prior work in [30]) to compute the near-optimal path for the characterization vehicle. Suppose that the complexity of computing the characterization path when cell-wise characterization gains are known is similar to that of the search path.…”

“…where η \ q i denotes the set of cells in η except cell q i . The terms up to the third summation in (30) denote the characterization gain of acquiring the environment measurement y from cell q i ∈ η, and the other terms that start with the third summation denote how likely it is that sampling cell q i will improve the performance of a follow-on search mission. That is, it represents the chances that cell q i will be visited during a follow-on search mission based on the environment measurements that we may acquire along the path η.…”