This is a partial list of research highlights for the Navy Center for Applied Research in Artificial Intelligence. Click on a highlight title to see more details about the research.

Participant uses NRL's experimental multimodal, multitask testbed. 3D Audio-Cued Operator Performance Modeling

This project, known as the "Auditory Principles" project, is basic research seeking to identify and model attentional principles for integrated uses of virtual audio information in Navy watchstanding operations. Navy watchstanding operations increasingly involve environments in which operators must attend to more than one critical task display at a time. In response, the Navy is pursuing a model-based understanding of human performance in multitask settings.

Picture of simulation Adaptive Testing of Autonomous Systems

The objective of this project is to develop technologies for advanced test and evaluation of the control software for intelligent autonomous systems. Machine learning techniques in the form of evolutionary algorithms and reinforcement learning are applied to learn the minimal number of faults that cause minimal or failed behavior of an autonomous system which is under the control of autonomy software -- the autonomy software is the system under test. The method uses high fidelity simulations of the vehicle and the environment, and tests are performed in that simulation.

A Tactical Action Officer (TAO) interacts with computer displays and audio inputs. Chat Attention Management for Enhanced Situational Awareness

This applied research project focuses on novel techniques and learning algorithms for (1) classifying chat messages, 2) summarizing chat room situations, and (3) 3D audio and visual cueing techniques for intelligently cueing watchstanders in a Navy Combat Information Center (CIC).

Octavia depicted in maintenance task where good cognitive skills are important Cognitive Robotics and Human Robot Interaction

More capable and intelligent robots and autonomous systems will require more human-like cognitive abilities. The hypothesis is that systems that use human-like representations, strategies, and knowledge will enable better collaboration between these autonomous systems and the people who use them. An autonomous system must be able to perceive its environment, explain its decisions, and predict a person's needs,which should lead to more trust in the autonomous system. In this line of research, computational cognitive models, such as ACT-R, are used to build process models of human cognitive skills, and those models are then used as as reasoning mechanisms on the autonomous systems.

Decision tree graphically depicting decision processes in a law enforcement scenario. Cognitively Inspired Decision Making for Visualization

The major objective of this project is to investigate the potential advantages of using a cognitively-based approach to autonomous decision making at multiple levels in a command structure. The intent of this approach is to facilitate the presentation of autonomous reasoning to human decision makers in ways that allow them to rely on and/or revise decisions that have been made at lower levels. This research will promote rapid situational understanding of the battle space and facilitate the decision maker’s ability to intervene and override the system as needed.

Octavia the robot assisting in extinguishing a fire. Damage Control for the 21st Century

The DC-21 project is developing advanced technologies for shipboard damage control. The project includes the development of advanced autonomous systems to assist in damage control and discovery and control of incipient fires. Our focus is specifically on the human-robot interaction technology that will allow a Navy firefighter to interact peer-to-peer, shoulder-to-shoulder with a humanoid robotic firefighter.

Goal-Directed Autonomy Goal Reasoning

Autonomous agents should be able to identify what goals they should pursue at any time during their execution by reasoning about possible objectives, their observations, and their environment model. Our investigations on goal reasoning concerns the design, implementation, and evaluation of agents with this capability (e.g., for controlling an unmanned vehicle).

Sonobuoys are used to make an impulsive broadband signal (essentially an explosion), and its reflections (echoes) are evaluated to determine if potentially dangerous underwater objects of interest (subs, mines, etc.) are in the vicinity. Human Mimetic Active Sonar Classification

Exploratory development in this research involves the development of algorithms, based on human perception of sound, for improved and automated active sonar classification. Single-ping algorithms transform sonar sounds into aural features for a new classifier that classify sounds.The project uses archived data sets and human subject research results to develop human-mimetic transforms to generate set-theoretic features to describe sounds. The project seeks to develop classifiers based on these types of features.

Phonological Analysis of Speech on Hand-held Smartphones produces langugage/dialect identification and associated confidence levels. Machine Classification of Spoken Language

Machine Classification of Spoken Language is basic research involving the classification and identification of the speech patterns of native and non-native speakers of English using phonetic and phonological analyses. The initial phase of this research involved the hand-coding of phonetic and phonological rules to determine speakers' native accents from passages spoken in English. The research progresses into the Automatic Dialect Identification project that provides machine-driven phonological analyses with confidence levels to identify the native language/dialect of both native and non-native speakers of English. Machine learning techniques will be used to train the phonological analysis engine in order to automate the procedure.

Diagram showing interaction on platforms, mission and algorithms Mobile Autonomous Navy Teams for Information Surveillance and Search

The focus of this effort is to adapt techniques inspired by nature to information gathering in complex Navy environments. This requires search techniques that are efficient when possible and robust when necessary, a level of adaptability that is common in nature and thus suggests a biologically-inspired approach. We refine the assumptions the algorithms make about the mission environment and platforms they direct in order to take a systems engineering approach to advancing the state of the art of robotic information gathering.

Change points or naturally occurring breaks in accelerated speech affect comprehension. Naturally Occurring Change Points in Navy Radio Communications

The project addresses the problem of monitoring multiple radio communication channels. By acceleration and serialization of radio traffic from concurrently active channels. Using these techniques, Navy watch standers have an increased comprehension of the communication channels, and miss less communications. Building on related work being conducted at NRL, a series of experiments are designed to determine to what extent the skill of attending to and understanding accelerated speech can be trained, and whether or not individual differences in memory and executive factors contribute to task performance in complex auditory domains.

ROC Curve for error model Predicting and Preventing Errors

Even when people know how to do a task very well, errors still occur: People forget to check their blind spot when driving or they leave their original on the glass after making a copy. Our goal is three-fold: to understand why people make errors, to build models that take that understanding and predict when someone is going to make an error, and then to prevent that error before it happens.

Tactile sensors attached to an MDS robot Robotic Touch Sensing, Manipulation, and Fault Detection

The objective of the Robotic Touch Sensing project is to develop an artificial sensate skin for robots to extend the perceptual capabilities of robotic manipulators to include touch. Under this effort we are developing tactile sensor arrays using piezoresistive sensing elements and have demonstrated a method for determining the location and magnitude of a contact (or contacts) for a multi-touch artificial skin by analyzing the responses from the sensors embedded within the skin.

Diagram of physicomimetic control law based on gravity. Swarm Control using Physicomimetics

Swarm intelligence is characterized by the emergence of collective capabilities from simple autonomous agents resulting from local interactions between the agents and their environment. Natural examples of swarm intelligence (e.g., colonies of ants) have led to the development of a number of distributed approaches to controlling agents. The method of swarm control we will use in this project is called physicomimetics. This method is based on an artificial physics representation in which agents behave as point-mass particles and respond to artificial forces generated by local interactions with nearby particles. We have developed a generalized form of physicomimetics that supports heterogeneous agents through multiple particle types and multiple force laws.

Example result. Trinocular Structured Light System

TSLS is a project motivated by the need to provide real time protection of moving convoys from attack by ambush with small arms. Our approach is to develop a 3D machine vision system which can detect the weapons by shape. Rocket propelled grenades (RPGs) and similar weapons are a great threat to our forces, in part because of their low cost and easy availability to our asymmetrical adversaries.

Unifying Inference through Attention Unifying Inference through Attention

This research program has as its goal the development of a cognitive system that acquires strategies for controlling inference. Much like humans can learn to solve mathematical equations, prove logical theorems, analyze filmic metaphor, and construct legal arguments, a broadly intelligent system must be able to develop new forms of reasoning about the world. We claim that the manipulation of attention, directing it toward specific thoughts and perceptions, provides a way to carry out different kinds of reasoning. Moreover, learning attention strategies involves attending to trains of thought that make sense of the world.