Intent Recognition for On-Water Adversarial Agents
Abstract
The problem. This project addresses the problem of detecting and identifying adversarial intentions performed by individual or group,s of vessels (red) that pose threats to an own (blue) ship. In prior ONR-funded research, we developed Hidden Markov Model (HMM)-bas,ed capabilities for recognizing intentions of single agents and we evaluated the system using simulated data for both training and v,alidation. A small scale validation using physical boats has been performed by HII Unmanned Systems (HII) in 2020. The results of th,e validation indicate the feasibility of the HMM approach while showcasing the limitations of simulation environments and of control,ler-based navigation behaviors. In this project, we propose to establish a tight collaboration with HII to further develop and refin,e our intent recognition capabilities for validation on physical unmanned surface vessels (USVs).Proposed solution. We propose an in,teractive, tight-loop coordination between UNR and HII in which the intent recognition capabilities will be continuously developed a,nd refined at UNR based on feedback received after testing and evaluation on USVs by HII. HII will assist by providing the following,: 1) subject matter expertise regarding realistic, human-led navigation maneuvers, along with intentions and scenarios of interest,,2) data for individual navigation behaviors acquired from USVs, and 3) textual descriptions of and data for complex navigation scena,rios that involve successions of several maneuvers as well as adversarial attacks by multiple vessels. At UNR, we will investigate t,he following research problems: 1) design of different representations for the navigation behaviors (lower-level of modularity, hier,archical HMMs), 2) detection of intentions by adversarial groups, and 3) intent recognition under passive (blue ship follows prescri,bed plan, does not take evasive maneuvers) as well as active (blue ship takes evasive maneuvers) response policies.Research objectiv,es. We will work toward the following specific research objectives:RO-1. Acquire expert knowledge and USV data from HII for realisti,c on-water navigation.RO-2. Continuously develop, (re-)train and refine behavior models using feedback from HII.RO-3. Develop capabi,lities for detecting group and deceptive intent.RO-4. Develop a validation methodology of the intent recognition system, jointly wit,h HII.Approved for Public Release.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 05, 2022
- Source ID
- N000142212176
Entities
People
- Monica Nicolescu
Organizations
- Nevada System of Higher Education
- Office of Naval Research
- United States Navy