Advanced Capabilities in Multi-Agent Search for the HAUV Ship-Inspection Vehicle
Abstract
Executive Summary The Office of Naval Research (ONR) has been investing in fundamental research in feature-based navigation (FBN) with an eye toward improved accuracy of autonomous navigation in the noncomplex and complex areas of a ship during ship-hull searches performed by a Bluefin Hovering Autonomous Underwater Vehicle (HAUV) (Fig. 1). The metrics by which the autonomous search improvements are measured include search rate, target re-acquisition accuracy, and percentage of ship searched. We believe that by adding a second vehicle to these demonstrations we can begin to look at fundamental problems associated with new challenges in multi-agent operations such as collaborative FBN and multi-agent planning—moving beyond single vehicle performance measures to more how a vehicle works in a collaborative method with other unmanned underwater vehicles (UUVs) and control centers in the operating area. The new metrics that would be tracked would include interoperability and reduced mission time. The ability to share FBN-derived maps and target waypoints between two vehicles will aid in response times by explosive ordnance disposal (EOD) teams and aid in information sharing when working with vehicles of another class, which will help the overall mine countermeasures (MCM) and very shallow water (VSW) missions. Figure 1: The Bluefin Robotics HAUV is specifically designed for in-situ underwater hull and harbor inspection and requires no pre-existing infrastructure (e.g., acoustic beacons) to localize. Instead, it uses a DVL and FBN for hull-relative navigation, which differentiates it from other similar inspection robots. To accomplish this, we propose new and continued work in advanced capabilities and extensions of the HAUV navigation and perception systems, to be carried out in collaboration between the University of Michigan (UM) and Carnegie Mellon University (CMU), over a period of three years. These advanced capabilities build upon our foundational work in single-vehicle FBN and planning for inspection capabilities, and seek to yield integrated methods in multi-agent, automated, whole-hull search. Current work has demonstrated large-scale, repeatable, multi-session FBN performance in non-complex area (NCA) regions of a hull environment, as well as automated path planning for 100% sensor coverage in the complex area (CA) screw/rudder regions, as well as the ability to perform whole-hull search by fusing the CA and NCA surveys. The proposed work will expand these efforts to include FBN-based navigation inspection of non-vessel marine structures (e.g., pier); sharing of FBN-derived maps between multiple vehicles; and transfer learning of FBN maps acquired under different modalities (e.g., acoustic and optic). The proposed work is in collaboration between UM (Eustice) and CMU (Kaess) with a demonstration plan involving Naval Explosive Ordnance Disposal Technology Division (NAVEODTECHDIV) (Kick) and the requested use of two government furnished equipment (GFE) HAUV vehicle assets. 3
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 03, 2016
- Source ID
- N000141612103
Entities
People
- Michael Kaess
Organizations
- Massachusetts Institute of Technology
- Office of Naval Research
- United States Navy