We describe preliminary work toward applying a goal reasoning agent for controlling an underwater vehicle in a partially observable, dynamic environment. In preparation for upcoming at-sea tests, our investigation focuses on a notional scenario wherein a autonomous underwater vehicle pursuing a survey goal unexpectedly detects the presence of a potentially hostile surface vessel. Simulations suggest that Goal Driven Autonomy can successfully reason about this scenario using only the limited computational resources typically available on underwater robotic platforms.

Motivated by the predictions of a theoretical model developed to describe the acoustic flow force exerted on closely spaced nano-fibers in a viscous medium, we have demonstrated a novel concept for a particle velocity-based directional acoustic sensor. The central element of the concept exploits the acoustically induced normal displacement of a fine mesh as a measure of the collinear projection of the particle velocity in the sound wave. The key observations are (i) the acoustically induced flow force on an individual fiber within the mesh is nearly independent of the fiber diameter and (ii) the mesh-flow interaction can be well-described theoretically by a nearest neighbor coupling approximation. Scaling arguments based on these two observations indicate that the refinement of the mesh down to the nanoscale leads to significant improvements in performance. The combination of the two dimensional nature of the mesh together with the nanoscale dimensions provides a dramatic gain in the total length of fiber exposed to the flow, leading to a sensitivity enhancement by orders of magnitude. We describe the fabrication of a prototype mesh sensor equipped with optical readout. Preliminary measurements carried out over a considerable bandwidth together with the results of numerical simulations are in good agreement with the theory, thus providing a proof of concept.

This paper studies the problem of autonomous data collection where an underwater vehicle is required to reach several target regions within a specified time limit. The proposed approach takes into account the vehicle dynamics, the time-varying ocean currents, and the obstacles in the region in order to effectively plan a collision-free and dynamically feasible trajectory whose time duration does not exceed the time limit. When the time limit makes it impossible to reach every target, the approach seeks to reduce the penalty accrued by the target regions that are not visited. The approach combines sampling-based motion planning with constraint-based solvers. In fact, a constraint-based solver searches a navigation roadmap to compute bounded tours which minimize the accrued penalty. Sampling-based motion planning is then used to expand a motion tree along these tours. Unsuccessful tour expansions are penalized to promote exploration of alternative tours. Simulation and field experiments demonstrate the efficiency of the approach in planning collision-free and dynamically feasible trajectories that reduce the accrued penalty.

This paper presents an efficient approach for planning collision-free and dynamically feasible trajectories that enable a mobile robot to carry out tasks specified as regular languages over workspace regions. A sampling-based tree search is conducted over the feasible motions and over an abstraction obtained by combining the automaton representing the regular language with a workspace decomposition. The abstraction is used to partition the motion tree into equivalence classes and estimate the feasibility of reaching accepting automaton states from these equivalence classes. The partition is continually refined to discover new ways to expand the search. Comparisons to related work show significant speedups.

This paper seeks to enhance the autonomy of underwater vehicles. The proposed approach takes as input a mission specified via a regular language and automatically plans a collision-free, dynamically feasible, and low-cost trajectory which enables the vehicle to accomplish the mission. Regular languages provide a convenient mathematical model that frees users from the burden of unnatural low-level commands and instead allows them to describe missions at a high level in terms of desired objectives. To account for the constraints imposed by the mission, vehicle dynamics, collision avoidance, and the complex spatial and temporal variability of the ocean environment, the approach tightly couples mission planning with sampling-based motion planning. A key aspect is a discrete abstraction obtained by combining the finite automaton representing the regular language with a navigation roadmap constructed by probabilistic sampling. The approach searches the discrete abstraction to compute low-cost and collision-free navigation routes that are compatible with the mission. Sampling-based motion planning is then used to expand a tree of dynamically feasible trajectories along the navigation routes. The approach is validated both in simulation and field experiments. Results demonstrate the efficiency and the scalability of the approach and show significant improvements over related work.