United States

AI-Enhanced Undersea Signal Processing for Long-Range Vessel Classification

The ability to detect and classify surface vessels at long ranges—beyond visual and radar limits—is critical for distributed maritime operations and decision superiority. Acoustic sensing platforms offer unique advantages for this mission, yet classification remains challenging due to highly non-stationary signals, pervasive environmental clutter, and limited labeled datasets. The Office of Naval Research (ONR) has identified this capability as a research priority. Avathon AI proposes an innovative approach that fuses physics-based modeling with advanced machine learning to deliver robust performance in complex acoustic environments.
Our framework integrates multiple technical components:
• Physics + AI Hybrid Models to improve acoustic imaging and classification beyond traditional statistical methods.
• Mass Feature Construction (MFC) across MFCC, spectral, wavelet, and time-domain features to identify predictive patterns at scale.
• Transformer-Based Architectures, including Masked Spectrogram Modeling (MSM-MAE), for unsupervised labeling and resilient embeddings in cluttered environments.
• Data Augmentation leveraging large public acoustic datasets to bootstrap models for naval-specific signatures.
This approach enables rapid adaptation to new vessel classes with minimal labeled data, improves classification accuracy under distortion and interference, and supports operational decision-making at the tactical edge. During the proposed 20-minute session, Avathon will demonstrate feature construction workflows, transformer-based classification on representative datasets, and discuss transition pathways for ONR and NIWC Pacific programs focused on ISR, undersea warfare, and multi-domain awareness.
Avathon AI brings over a decade of experience delivering AI/ML solutions for government and industry, including recent acoustic projects with NSWC. Our proven track record in predictive maintenance, ISR fusion, and autonomy platforms positions us as a strong partner for advancing naval AI capabilities.

NAML 2026

AI-Enhanced Undersea Signal Processing for Long-Range Vessel Classification

niwc pacific.

acoustic classification

undersea signal processing

machine learning

transformers

### Welcome to the 10th annual workshop on Naval Applications of Machine Learning!

NAML is the premier annual event showcasing current machine learning research relevant to Naval applications. We welcome participation from the defense community, industry, and academia. The NAML workshop is an opportunity for researchers from Navy research and development centers and our partner organizations to exchange technical ideas and to come together as a team. 
Naval Information Warfare Center Pacific is partnering with AFCEA San Diego chapter to host the workshop.
#### On-Demand Access

* This is an on-demand site and all the recorded content from the workshop is available here. 
* Access to the recordings is exclusively available to registered participants.
* Each registrant has received an email notification, with instructions on how to access the content.

[![](https://assets.underline.io/markdown_image/1/image/60cd12aa376c191d3467ea2ee60944ea.png)](https://naml2026.org/)

The goals of NAML are:

* To spread awareness of current machine learning research relevant to Naval applications
* To connect machine learning researchers with experts in military requirements
* To build and strengthen collaborations within the defense research community

#### *"What we do as a Navy is a human business. Don’t pass up an opportunity to come together as a team. Don’t pass up leadership opportunities, and don’t outsource them to a PowerPoint. If you have an opportunity to come together as a team, to increase belonging, do it."*
Admiral John M. Richardson, 31st Chief of Naval Operations
<br><br>

This is an on-demand site. All recorded content from NAML 2026 is available here and access to it is exclusively available to registered participants.

**Please log in using the email address you registered with.** If you have any issues logging in, please contact us at <naml2026@underline.io>.

Let’s celebrate the past decade by investing in the next. 🚀


Please log in to explore this event.

This is an on-demand site. All recorded content from NAML 2026 is available here and access to it is exclusively available to registered participants.

**Explore the latest advancements in machine learning tailored to Naval applications 🔎**

With on-demand access to NAML 2026, you can watch all lectures & browse poster presentations at your convenience, while also engaging directly with presenters via the Underline platform - ask questions, exchange ideas, and stay connected.

**If you still haven’t registered, you can do so now.**
👇
**Take advantage of a special one-time offer: full access for just $99, with complimentary registration for students, government, and military employees.**

**Get access here: https://naml2026-ondemand.paperform.co/**

Please check your spam folder, or contact our support team at **<naml2026@underline.io>** if you registered, but still haven't received an access email.

Let’s celebrate the past decade by investing in the next. 🚀

Please register!

The 10th Annual Workshop on Naval Applications of Machine Learning (NAML 2026) was held in San Diego, California, March 2-5, 2026. NAML is a forum for interdisciplinary experts from government, industry, and academia to discuss results of ongoing research on theory and applications of machine learning to address some of the most challenging problems faced by our naval forces. 



technical paper

The maritime industry relies heavily on the Automatic Identification System (AIS) to enhance navigational safety and ensure effective vessel traffic management.[1] AIS is a tracking system that enables ships to communicate their position, speed, heading, and other vital data with one another and with shore-based authorities.[1] However, the efficacy of AIS can be severely compromised when anomalies in the data occur.[2] These anomalies can arise from various sources, including equipment malfunctions, data entry errors, environmental interferences, and intentional malicious actions such as data spoofing.[2]

Such anomalies pose significant risks to maritime safety, security, and operational efficiency, leading to potential accidents, navigational errors, and even legal issues for vessel operators.[2] For instance, an undetected AIS anomaly may result in a vessel being unaware of a nearby ship, increasing the risk of collision. Moreover, anomalies may mask illegal activities such as unauthorized fishing, piracy, or human trafficking, challenging law enforcement and regulatory agencies tasked with maintaining maritime order. The failure to detect these anomalies in real-time can have dire consequences for both vessels and marine ecosystems, highlighting the critical need for robust anomaly detection solutions.[2]

Despite the importance of effective AIS monitoring, traditional methods of data analysis struggle to keep pace with the sheer volume and velocity of maritime traffic data generated daily.[1] As the number of vessels in commercial shipping increases, so does the complexity of data interpretation.[1] [2] To address these significant challenges, there is a growing trend toward integrating Artificial Intelligence (AI) and Machine Learning (ML) technologies into maritime safety frameworks.[2] AI and ML possess the capacity to analyze large volumes of AIS data in real-time, identifying irregular patterns and potential anomalies that may escape human attention.[2] For example, supervised learning models can be trained on historical AIS data to recognize the typical behavior of vessels in various maritime environments, allowing them to flag deviations as potential anomalies.
Moreover, AI methods can integrate various sources of information, such as weather data, historical traffic patterns, and even satellite imagery, to provide a contextual understanding of vessel behaviors, enriching the analysis and improving detection capabilities.[2] 

References: 

[1] Yang, D., Wu, L., Wang, S., Jia, H., & Li, K. X. (2019). How big data enriches maritime research–a critical review of Automatic Identification System (AIS) data applications. Transport reviews, 39(6), 755-773.

[2] Lee, E., Mokashi, A. J., Moon, S. Y., & Kim, G. (2019). The maturity of automatic identification systems (AIS) and its implications for innovation. Journal of Marine Science and Engineering, 7(9), 287.

Anomalies in Automatic Identification System (AIS) Maritime Traffic Data

Modern destroyers generate far more engineering data than humans can interpret in real time. Legacy maintenance models using planned and casualty maintenance supported by tools like ICAS (Integrated Condition Assessment System) struggle to recognize anomalies in the data streams or turn this deluge of data into actionable recommendations that integrate into sailor workflows.

This talk introduces ERMv4, the Navy's Enterprise Remote Monitoring system which turns roughly 7,000 hull, mechanical, and electrical sensor readings per second into recommendations that feed the ship’s existing maintenance planning system. This shift has moved condition-based maintenance (CBM) from concept to operational tool on USS Fitzgerald (DDG-62).

We focus on three themes that connect ML research with naval requirements and fielded impact: 

ML at the tactical edge. 

ERMv4 enables the MLOps data flow for models to learn from noisy, heterogeneous sensor data across pumps, motors, power distribution, and auxiliary systems. Delivering useful signals from tactical platforms demands time-series preprocessing, anomaly and degradation modeling when failure labels are rare, and handling non-stationary behavior after maintenance or configuration changes. We also cover edge-compute realities: packaging models for afloat deployment, operating under SWaP and bandwidth limits, handling DDIL connectivity, and safely updating models on ships underway for months within policy and accreditation constraints. 

Human-in-the-loop maintenance. 

ERMv4 is built so the computer does not “boss” sailors around. Recommendations are ranked, explained, and integrated into existing maintenance systems in ways that leverage sailor expertise and reduce alert fatigue. UI and workflow choices make it clear why a recommendation is being made, how confident the system is, and what evidence backs it up. We share early lessons on trust: what ship’s force accepts, what they push back on, and how the system must evolve so sailors feel supported, not treated like warehouse workers executing AI-generated task lists. 

Fleet-level implications. 

ERMv4 is one of the first AI programs-of-record continuously monitoring ship health and offers a template for scaling CBM across classes and platforms. We discuss how consistent data models, shared anomaly patterns, and common integration points with Navy logistics and maintenance systems can support Distributed Maritime Operations and higher availability, and how “assured autonomy” for CBM—demonstrating model behavior, reliability, and limits—can build durable trust in recommendations. 

Attendees will leave with a grounded case study of naval ML running on the deck plates, plus practical lessons for building and deploying ML systems that sailors actually use in the fight.

Machine Learning at Sea with Speed: ERMv4 and Condition-Based Maintenance for Mission-Ready Destroyers

Machine-learning enabled surrogates are increasingly used to emulate complex high-energy-density physics (HEDP) phenomena, yet the scarcity of high-fidelity, multi-material shock-propagation datasets remains a major barrier to advancing predictive modeling and inverse design capabilities for defense-relevant applications. The Perturbed Layered Interface (PLI) partition of the High-Explosives and Affected Targets (HEAT) dataset addresses this gap by providing time-resolved simulations of explosive-driven shocks interacting with heterogeneous materials. Within the PLI dataset, a fixed material suite subjected to varied initial interface perturbations produces a wide range of plastic deformation outcomes, enabling systematic study of geometry-dependent shock dynamics.

Using this dataset, we develop an inverse-design framework that reconstructs probable initial configurations given only the final, post-shock material morphologies. Our approach leverages machine-learning models trained on PLI simulations to infer a distribution of initial conditions consistent with an observed final deformation. These inferred distributions are subsequently validated through forward simulations that propagate the inferred initial states to late-time outcomes, demonstrating strong agreement with target morphologies. This bidirectional modeling capability provides a physics-informed pathway for linking final observable signatures to their causal shock-driven origins.

Our results highlight the utility of high-fidelity simulation datasets for learning complex multi-material shock behavior and demonstrate an end-to-end inverse design protocol relevant for verification of explosive-driven systems, assessment of material responses under extreme loading, and the development of ML-based decision tools for naval and defense applications.


Inverse Design of Shock-Driven Material Deformation Using Machine Learning and the PLI Dataset

A single agent represents a single system capable of ingesting local data, indexing, cataloging information, performing knowledge pattern discovery, and separating patterns and
anomalies from data. Multiple agents work collaboratively in a peer-to-peer network. Each agent has a peer list. Such multiple agents’ collaboration can be modeled as cooperative games. Each
agent optimizes its own objective locally. We show that each agent self-organizes or converges to its best value and the whole agent network achieves the best social welfare based on both the quantum adiabatic evolution transformation (QAET), and quantum intelligence game (QIG) or the QAET-QIG framework. 

The contributions of this paper are summarized as follows:
1) We show that a multi-agent quantum system converges to an equilibrium state which has a high degree of purity and coherence via a QAET-QIG evolution in a super-positioned torus. The convergence results in an equilibrium and measurable property or quantum theoretic value
(QTV) of the system that is optimized. We demonstrate the QAET-QIG framework and show two theorems, one algorithm used to score new data for quantum values, and compare them with the traditional QAET.
2) We apply the QAET-QIG framework to the kill web concept that potentially improve the traditional kill chain process such as find, fix, track, target, engage, and assess
(F2T2EA). The improvement is measured in the values of powerful global optimization, distributed lethality, and load balancing. 

We show a use case of the QAET-QIG framework to a potential application of mixed sensors, platforms, weapons, and effects.

Disclaimer: The views and conclusions contained in this document are those of the authors and should not be interpreted as others or representing the official policies, either expressed or implied of the U.S. Government.

Kill Webs by Collaborative & Self-organizing Agents (CSOAs)

GNSS is critical to accurate Positioning, Navigation and Timing (PNT) solutions in many naval applications, but is susceptible to jamming and interference. Therefore, methods to maintain a user’s position in the absence of consistent GNSS are highly sought after. We propose our method to address GNSS-denied pedestrian navigation, named the Machine-learning Aided gait Recognition for Inertial Navigation and Orientation (MARIO) system. MARIO uses three main components to estimate a user’s position and orientation: 1) microelectro-mechanical system (MEMS) IMU sensors mounted on different parts of the body, 2) a convolutional neural network (CNN) that estimates movement velocity based on the sensor’s raw IMU data, and 3) a Kalman filter which takes the sensor’s raw IMU and the CNN’s speed estimation as inputs, and outputs the user’s estimated position. In this presentation, we demonstrate the latest advancements in our fully developed real-time navigation system. These advancements include improved data-collection and processing methods that allow for dynamic movements, wireless communication integration and monitoring, as well as camera and phone-app integration. The MARIO system was able to achieve under 1% error over total distance traveled in fully GNSS-denied environments (USS Midway). We will showcase detailed results of our system’s performance, including photos of our demonstrations conducted over the previous year.

GNSS-Denied Pedestrian Navigation in Real-Time using CNNs

Large Language Model (LLM) agents increasingly rely on external tools to accomplish complex tasks, and how they scaffold tool invocations can influence both efficiency and reliability of the agent. 

The classic Model Context Protocol (MCP) requires an LLM agent to emit a structured JSON “tool‑call” that is parsed by an intermediary before execution. Code‑Mode, by contrast, lets the LLM write short programs that call typed MCP bindings inside a sandboxed runtime; because models are trained on extensive code corpora, this approach may allow agents to compose multiple tool calls more naturally and with fewer tokens. 

A second dimension—inter‑agent communication—has received comparatively little systematic study. In hierarchical or multi‑agent settings, an agent often delegates subtasks to subordinate agents that in turn invoke tools. We examine three inter‑agent communication modalities: 

- Direct: Code‑Mode calls between agents’ MCP bindings. 
- Classic: MCP messages that encode tool requests. 
- Natural‑language: queries that the receiving agent must interpret before acting. 

To compare these modalities, we developed a controlled evaluation measuring three criteria: 
1. Token efficiency — tokens consumed per completed workflow step. 
2. Reliability and error rate — frequency of malformed calls or execution failures. 
3. Coordination overhead — number of round‑trips and latency incurred by the communication method. 

This poster will present a systematic benchmark for assessing whether Code‑Mode’s theoretical advantages translate into practical gains in token efficiency and inter‑agent coordination. Both Code‑Mode and MCP are publicly documented; researchers can replicate our framework to evaluate alternative tool‑invocation strategies in their own hierarchical LLM agents.

Code‑Mode vs MCP: Token Efficiency and Inter‑Agent Communication in Hierarchical LLM Agents

Defense operations increasingly demand effective human-AI teaming under conditions that severely strain cognition: simultaneous information streams, time-critical decisions with incomplete data, and sustained operations inducing fatigue and stress. Current AI systems face a critical limitation: they model operator behavior without accounting for how cognitive state affects that behavior. When an intelligence analyst revisits the same data repeatedly or a UAV operator exhibits degraded response times, existing systems cannot distinguish intentional strategy shifts from cognitive overload. This ambiguity undermines AI assistance—systems may present additional information when operators lack capacity to process it, or fail to intervene when fatigue compromises decision quality.

We have developed HiVAE, a hierarchical variational autoencoder inspired by the belief-desire-intention structure of human Theory of Mind (accepted to AAAI 2026 Theory of Mind Workshop). By structuring latent variables to mirror the causal dependencies in cognitive ToM models—where beliefs inform desires, which inform intentions—our three-level BDI hierarchy substantially outperforms non-hierarchical baselines on goal inference tasks, achieving significantly lower prediction error and maintaining robust performance when agent preferences shift. Ablation studies confirm this performance advantage is directly attributable to the hierarchical BDI organization itself. This work demonstrates that embedding ToM-motivated causal structure into deep architectures yields measurable gains in behavioral prediction. Our ongoing work extends this foundation by integrating physiological monitoring to capture how affective states modulate the mental states we can now infer.

This capability enables a critical extension for operational contexts: Affective Theory of Mind (AToM), which integrates continuous physiological monitoring to model how stress, arousal, and fatigue modulate inferred mental states. By incorporating biosensor data (heart rate variability, electrodermal activity, movement patterns) alongside behavioral observations, AToM would distinguish behaviors driven by mental state changes from cognitive resource depletion. We propose a hybrid architecture where deep learning models process physiological streams to generate affect estimates, which condition large language model inference of mental state hypotheses for Bayesian inverse planning. This would enable AI teammates that adapt support to operator cognitive capacity—recognizing when narrowed analyst focus reflects overload rather than strategy, or when degraded UAV operator performance signals need for handoff rather than increased autonomy.


Affective Theory of Mind: Physiologically-Informed Mental State Inference for Human-Machine Teaming

This project develops a system for remotely launching unmanned aerial vehicles (UAVs) and autonomously executing waypoint-based flight plans around a moving target. As the target object translates or rotates, the UAV continuously updates its trajectory by shifting each planned waypoint based on real-time target motion data. Target tracking is achieved either through GPS stations mounted on the target or via an in-house computer vision model capable of estimating target pose. The UAVs were custom-built in the RESTORE laboratory and the autonomy was implemented using ROS 2 with pymavlink communication. The GPS-based system has been successfully executed as two UAVs were remotely launched from a location hundreds of miles away and autonomously circumnavigated a designated target while capturing imagery at defined intervals and transmitting it to a remote server in real time. The computer vision–based tracker has been validated independently and is currently being integrated into the full pipeline.

Custom UAVs for Remote Autonomous Launch and Moving Target Tracking

As large language model (LLM)-powered software is given greater freedom to act autonomously via tool calling and agentic frameworks, the need to monitor its behavior in order to minimize unintended actions continues to grow. The Multimodal Sensemaking in SeaVision (MUSES) project at NIWC LANT has used several open-source tools to test and evaluate LLM's and associated tools that are part of our software. MUSES is also researching using a separate LLM framework to continuously monitor and analyze T&E results to enable autonomous self-correction when our systems fail or otherwise behave unexpectedly. This poster presents an overview of our T&E work using several publicly releasable case studies.

Open-source tools for LLM T&E in a maritime awareness domain

Autonomous systems need environmental cues for orientation, navigation, and specialized tasks. Interpreting the layout of a space can be done by detecting objects and reading signs, which lead to enhanced scene understanding like identifying and assessing equipment in large spaces. We propose a pipeline for detecting, reading, and documenting relevant signs within a ship to aid in shipboard navigation and inventory tasks. This pipeline distinguishes between important Navy shipboard signage, including placards, exit signs, configuration labels, and bullseyes. Once the signs have been detected and classified, they are fed into specialized optical character recognition algorithms to read and decipher text according to the automated task. Finally, the sign classes and transcriptions are corrected via post-processing and mapped into a corresponding point cloud to form an inventory of the target space. This unique, proposed system comprises algorithms to be used in-situ for a set of inventorying and navigation tasks.

Premium content

Next from NAML 2026

Anomalies in Automatic Identification System (AIS) Maritime Traffic Data

Similar lecture

Avathon AI-Advanced Decision Making