United States

SOAR (Smart Obstacle Avoidance through Reinforcement learning) presents a novel, adaptive framework for robust collision avoidance in multi-rotor aerial vehicles operating in dynamic, unpredictable environments. Traditional obstacle-avoidance systems for drones often rely on static rule sets, hand-tuned controllers, or assumptions about environmental regularity. These approaches struggle when confronted with rapidly evolving conditions such as sudden wind gusts, shifting obstacles, or the presence of other aircraft. SOAR addresses this limitation by training an end-to-end reinforcement learning (RL) agent capable of continuously adjusting its control strategies in response to real-time disturbances while maintaining adherence to a prescribed flight path.

The proposed system is developed and evaluated using the NVIDIA Isaac Sim environment, which enables high-fidelity simulation of quadcopter dynamics, sensor noise, environmental chaos, and multi-agent interactions. This platform allows the training pipeline to expose the agent to a broad distribution of challenging scenarios that would be impractical or unsafe to reproduce in real-world testing. Environmental perturbations include wind fields, transient forces, unexpected moving obstacles, and simulated aircraft exhibiting independent motion models. By integrating these diverse disturbances into training, SOAR emphasizes resilience, generalization, and situational awareness.

A key innovation of SOAR lies in its unified reward structure that jointly penalizes collisions, path deviation, and unstable control actions while positively reinforcing precise trajectory following under chaotic conditions. Rather than merely avoiding obstacles, the agent must intelligently compensate for external forces and maintain stable flight, transforming the problem from simple reactive avoidance to proactive, context-aware navigation. This encourages the emergence of control behaviors that dynamically balance safety, efficiency, and path fidelity.

The resulting system demonstrates the potential for RL-driven flight controllers to outperform conventional approaches in scenarios where environmental conditions cannot be fully predicted or modeled. SOAR contributes a scalable methodology for training robust aerial autonomy and highlights the value of simulation-rich RL as a design paradigm for next-generation drone navigation systems.

NAML 2026

SOAR: Smart Obstacle Avoidance through Reinforcement learning

object avoidance

aircraft

reinforcement learning

Naval Information Warfare Center Pacific is partnering with AFCEA San Diego chapter to host the workshop.

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.

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

**Watch this video to hear about the AFCEA purpose.**

<iframe width=580 height=700 frameborder="0" scrolling="no" src="https://screencast-o-matic.com/player/creleVVcvRh?v=6&ff=1&title=0&controls=1" allowfullscreen="true"></iframe>

Please log in to explore this event. You need to log in with the email address you registered with. Access credentials have been sent to your email. 

Please be sure to check your spam and other email folders if you do not see an email confirmation right away.

Please log in to explore this event.

Registration for this workshop has now closed. We appreciate your interest and hope to welcome you at a future event. If you’d like to hear about similar upcoming events, please check our website or join our mailing list.

Registration for this workshop has now closed.

NAML 2021

The annual workshop on Naval Applications of Machine Learning 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. 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 and to build and strengthen collaborations within the defense research community.


technical paper

### 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. 🚀


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. 



The mass deployment of Unmanned Systems (UxS) to the fleet burdens operators with increasingly complex decision-making demands that threaten their ability to conduct effective Command and Control (C2) mission management. The decision space required for UxS operations will expand exponentially as future missions scale in size, employ heterogeneous multi-domain platforms, and operate within Denied and Degraded Environments (D2E). Addressing these challenges requires unified, integrated C2 tools that are context-aware, objective-driven, and capable of synthesizing diverse autonomy-enabling technologies. No single capability can meet these needs in isolation; closing these gaps demands a cohesive toolchain that weaves together multiple models into a unified C2-focused solution.

We introduce the Decision Engine for Situational Context, Autonomous Replanning, Task Execution & Scheduling (DESCARTES), a framework designed for rapid integration of advanced autonomy models supporting C2 mission management. DESCARTES incorporates optimized path planning, dynamic environment discretization, priority task queuing, agentic Artificial Intelligence (AI), large language models (LLM), and reinforcement learning (RL) models in a modular architecture.

DESCARTES consists of three core components: the event cloud, the datacore, and an advanced discretizer and path planner called Multi-resolution Adaptive Graphs for Intelligent Cellular Automata Route Planning (MAGICARP). The event cloud, inspired by the Air Force Research Laboratory’s Cognitively Enhanced Complex Event Processor (CECEP), uses MQTT to implement a trigger–response mechanism in which specialized topics activate event sequences defined by behavior models. The datacore functions as a behavior model that records event triggers into a local database, automatically generating new Postgres tables and entries when previously unseen event types are introduced.

MAGICARP is a multi-resolution environment discretization and path-planning framework that partitions the operational space according to local conditions while preserving global structure. Drawing on cellular automata principles, MAGICARP represents the environment as interacting hexagonal cells whose local updates shape higher-level topology. An entropy-guided refinement mechanism directs computational attention to regions of high informational change, enabling localized updates that support scalable, high-fidelity route planning in time-critical domains.

Together, these components enable DESCARTES to evaluate C2 mission plans against behavior models that provide contextual understanding, examine, feasibility, capability, and adherence to rules, doctrine, and commander intent. DESCARTES can integrate this data and replan missions, bring attention to constraint and restraint violations, and provide contextually aware decision support to C2 operators.

Decision Engine for Situational Context, Autonomous Replanning, Task Execution & Scheduling

Naval installations and waterfront infrastructure depend on rapid, repeatable geospatial analysis to inform operations, maintenance, and resilience. Today, those workflows are fragmented: LiDAR is processed in one system, elevation models in another, and change detection and data publishing often require multiple tools and manual coordination. The result is latency, analyst bottlenecks, and limited scalability for mission-critical decisions.

This presentation introduces GeoAgent, an agentic geospatial orchestration framework that automates multi-sensor analysis pipelines through natural-language tasking while preserving existing tools and data environments across DoD and Navy operations.

GeoAgent acts as an orchestration layer over standard GIS and remote sensing platforms (e.g., ArcGIS, ENVI, QGIS) and connects to data services such as STAC, OGC, and AGOL. Analysts can describe an operation in plain language, for example requesting the latest LiDAR or imagery over an area of interest, performing analysis, and publishing results, and GeoAgent automatically executes the required steps behind the scenes. The system uses a protocol-driven integration model (MCP-style service endpoints) to call analytic toolchains, manage parameters, and maintain full traceability for audit and reproducibility.

The approach has been demonstrated through multiple proof-of-concept use cases, including flood detection, infrastructure impact mapping, and LiDAR-based terrain assessment. Across these examples, GeoAgent reduces manual workflow time from hours to minutes, eliminates tool handoffs, and allows both GIS experts and non-technical users to run complex geospatial analyses through simple, conversational commands.

Operationally, GeoAgent supports faster mission response, automated terrain and asset updates, and real-time situational awareness for infrastructure and range management. It provides a scalable way to maintain geospatial currency and deliver consistent, repeatable analysis across mission areas while integrating within existing government toolchains and IT policies.


Agentic Geospatial Orchestration for Multi-Sensor Terrain Modeling and Rapid Infrastructure Assessment

This abstract outlines a framework for developing intelligent, interpretable control systems by distilling policies from Reinforcement Learning (RL) agents into instruction-tuned Large Language Models (LLMs). We leverage navy-relevant, Gymnasium-compatible simulators, such as gym-sailing, as an initial training testbed. While this specific environment models foundational tasks such as upwind navigation and tacking, the underlying methodology is designed to be scalable and directly applicable to a wide range of more complex, high-stakes naval control systems, including autonomous underwater vehicle navigation and dynamic multi-vessel coordination. We first train an RL agent to act as an "oracle," generating an optimal dataset of state-action trajectories. This data is then used for policy distillation via supervised fine-tuning (SFT) of advanced, instruction-based LLMs, such as Llama 3.1 and Qwen3 of different sizes and quantization, transforming them into specialized reasoning agents capable of real-time, closed-loop control.

The advantage of the resulting LLM-based controller is its inherent interpretability. Unlike the opaque RL agent, the LLM can articulate a "chain-of-thought" rationale for its control decisions, such as explaining why a specific rudder angle was chosen in relation to wind direction, boat speed, and the target location. These reasoning agents offer high-level insights into the system's functioning, provide actionable strategic recommendations, and excel at generalization. They can adeptly handle dynamic priority switching in multi-objective settings—such as re-tasking from "maximum speed" to "maximum comfort" or "avoid new obstacle"—by processing these new objectives as natural language instructions, a feat unachievable by traditional RL agents without costly retraining.

This capability has direct and powerful applications for naval personnel training. By integrating these reasoning agents into simulators, trainees can engage in an interactive learning loop, querying the AI "skipper" or "tactical officer" for its strategy. This transforms the simulator from a simple practice tool into a dynamic, "explain-by-example" instructional platform, significantly accelerating skill acquisition & deep strategic understanding.

This RL-to-LLM distillation framework is not merely theoretical; it has been successfully validated in other high-stakes, complex domains. As a key proof point, we applied this method to distill policies from a sophisticated multi-agent RL controller responsible for the end-to-end liquid cooling management of HPC systems, such as the Frontier supercomputer. In that case, the distilled LLM controller consistently matched the performance of the original RL policy across key metrics. This approach can also be proven effective in a host of classic OpenAI Gym control environments, demonstrating a robust and generalizable pathway for creating high-performing, interpretable agents from optimized, black-box policies.

Interpretable Naval Control: Reinforcement Learning Policy Distillation for Reasoning-based LLM Agents

The military is increasingly overwhelmed by the sheer volume and complexity of data generated by modern warfare. Traditional methods of processing and analyzing this information are often inadequate, leading to delayed decision-making, reduced situational awareness, and increased risk to personnel. To address these challenges, the military requires advanced AI systems that can efficiently process vast amounts of data, provide actionable insights, and automate routine tasks, enabling faster, more informed, and more effective decision-making. 

Raytheon has been conducting Independent Research & Development to address these challenges as part of our Cognitive Aids to Sensor Processing, Exploitation, and Response (CASPER™) initiative. 

As part of CASPER™, we are exploring cutting-edge Large Language Models (LLMs) and Agentic AI frameworks that can act as intelligent orchestrators and tactical assistants for C2ISR missions. These agents analyze vast amounts of data to deliver real-time alerts, task available sensors, and enhance situational awareness for mission operators. 

We would like to propose a presentation that examines areas where existing C2ISR solutions fall short, describes the approach we are taking with CASPER™ to address these challenges, and discusses ongoing research and development activities aimed at closing key technology gaps.

This presentation will include a short demonstration of CASPER™ utilizing Agentic AI to support a simulated C2ISR mission. Our demonstration will center on a mission scenario in the South China Sea, where a persistent orbit of MQ-4C Triton UAVs is tasked with wide-area surveillance. We will demonstrate how the Agentic AI system aids operators in detecting anomalous vessel activities and dynamically tasking sensors to gather additional information as required. Additionally, we will showcase how these Agentic AI systems enhance situational awareness by responding to natural language queries using the current operational picture and a comprehensive repository of information available through Retrieval-Augmented Generation (RAG).

This demonstration will utilize the RACIE simulation environment that we presented as a poster at NAML 2025. We are requesting a 20-minute Technical Talk for this demonstration, but we would be happy to present CASPER™ as a Lightning Talk or Poster as well.


CASPER™: Transforming C2ISR Systems into Intelligent Combat Team Members

Current United States Marine Corps (USMC) marksmanship and tactical assessments rely on disparate technologies and methods that often provide incomplete human performance data and/or delayed results. Furthermore, assessments often require a one-to-one instructor-to-shooter ratio, which limits throughput for basic marksmanship assessments and impedes measurement of tactical/force-on-force training at scale. Our current program is investigating novel solutions to these issues through the pioneering application of artificial intelligence/machine learning (AI/ML) capabilities embedded within tactical computer fabrics. Sensors (acoustic and accelerometry) and microcontroller units (MCUs) are strategically placed in a malleable fiber and woven into the fabric, creating an imperceptible wearable system with fully encapsulated data collection and performance analysis and prediction. The core of our methodology lies in the application of AI/ML models, including neural networks (NNs), distributed across the fiber’s MCUs. The AI/ML models are designed to analyze the complex factors that contribute to marksmanship performance, including shot timing, precision, and shooter movement. A pilot study focused on determining shot detection, using labeled acoustic data, found that individual tactical computer fibers on both left and right arms achieved high accuracy in shot detection of blind test data. The NN models were first trained on ~800 shots across eight shooters with different postures. The NN was composed of separable convolution and dense layers with ~4000 trainable parameters, small enough to be deployed within the MCU itself. A method of combining the global efficiency of shot detection by the NN and precise peak finding through Shannon energy algorithms was also demonstrated. Live-fire testing included USMC zeroing procedures, the Infantry Marksmanship Assessment course of fire, and complex live-fire and maneuver events. Across these events, the shot timing captured by the tactical computer fibers was within >95% of the data captured by the Joint Marksmanship Assessment Package, the current gold standard for marksmanship assessment in the USMC, and the AI/ML models achieved >93% accuracy in predicting shot detection, with a low rate of false positives. The models also achieved >91% accuracy in predicting shot precision based on pre-shot shooter movement data captured via accelerometry and validated via grouping measurement (mean radius), providing deeper insight into individual shooter lethality. Future research efforts will focus on several key areas to further enhance the capabilities of the tactical computer fabric system. We plan to expand the dataset used to train the AI/ML models, incorporating data from a diverse range of shooters with varying skill levels and experience. We will also explore incorporating additional physiological data measured by the fibers to further improve the predictive capabilities of the AI/ML models.

Artificial Intelligence/Machine Learning (AI/ML)-Enabled Tactical Computer Fibers for Measurement and Evaluation of Live-Fire Shot Timing and Precision

## Background
Wearable devices have become essential tools for continuous health monitoring, yet the physiological signals they collect exhibit substantial heterogeneity in sampling rates, waveform characteristics, sensor quality, and measurement configurations. These inconsistencies present significant challenges for learning robust and generalizable representations, particularly when deploying models across diverse health applications such as mental state recognition, body state inference, physiological parameter estimation, and disease risk screening. Although recent time-series foundation models have demonstrated strong general-purpose capability, they typically lack explicit architectural mechanisms to accommodate the multimodal and cross-sensor nature of wearable sensing data.

## Methods
To address these limitations, we propose NormWear, the first multi-modal foundation model specifically tailored for ubiquitous wearable sensing scenarios. NormWear introduces a channel-aware attention framework augmented by a shared liaison [CLS] token that facilitates both intra-sensor pattern extraction and inter-sensor relational modeling. This design enables the model to flexibly adapt to arbitrary sensor configurations and to capture meaningful cross-modal dependencies. We pretrain NormWear on a large and diverse collection of physiological signals, including PPG, ECG, EEG, GSR, and IMU, from multiple public datasets, using self-supervised objectives to encourage universal and transferable representations. The pretrained model is then evaluated under zero-shot, partial-shot, and full-shot settings across a comprehensive suite of downstream health-related tasks.

## Results
Extensive experiments on 11 publicly available wearable sensing datasets, covering 18 downstream applications, demonstrate the strong generalizability and robustness of NormWear. Across mental health assessment, body state inference, vital sign estimation, and early disease risk evaluation tasks, NormWear consistently outperforms state-of-the-art baselines. The model shows substantial gains in zero-shot transfer, maintains competitive performance with limited labeled samples in partial-shot scenarios, and sets new benchmarks in full-shot fine-tuning. Ablation studies further confirm the effectiveness of the channel-aware attention mechanism and the shared liaison token in enhancing cross-sensor adaptability.

## Conclusion
NormWear establishes a unified foundation model architecture for heterogeneous wearable sensing data, offering strong representational consistency across diverse physiological modalities and device configurations. By enabling reliable zero-shot and few-shot transfer to a wide range of health-related tasks, NormWear provides a scalable and practical pathway for building generalized wearable AI systems and advancing real-world digital health applications.

Toward Foundation Model for Multivariate Wearable Sensing of Physiological Signals

The Navy is working to streamline ship maintenance to achieve better fleet readiness and a “north star” of 75 mission-ready ships. This requires advanced analytics which leverage large maintenance datasets for predictive, proactive and preventative decisions. However, existing maintenance processes produce low quality data that prevents their use and limits insights from analytics. Efforts to improve this data through post-hoc curation have proven costly and time-consuming while failing to address the key problem: recording high-quality data in the first place. RUBICON (Rubric-Based Intelligent Curation) addresses this key problem by providing real-time AI-assistance and feedback that helps sailors adhere to data requirements as they record maintenance reports. Our prototype system focuses on 2-Kilo Maintenance Action reports (2Ks), which are written for every maintenance action taken on a ship and currently suffer from ubiquitous imprecise language, technical errors, and inconsistencies. RUBICON uses multiple agents and a large language model (LLM) to format, standardize, and organize text according to downstream data requirements and SME-informed best practices. This allows sailors of wide-ranging experience levels to produce expert-quality reports while entering simpler, unstructured descriptions of maintenance issues. As a result, sailors using RUBICON spend less time on report writing, and the reports they create support deeper insight. 

By improving the quality of data-rich 2Ks, RUBICON unlocks better analytics that will empower proactive decision-making. For example, improved equipment reliability estimates can empower Navy Program Executive Offices (PEOs) to choose contracts that reduce long-term labor and cost. There are several challenges for building our system in the 2K domain: (1) reports use language, terminology, and knowledge that is out-of-distribution for most LLMs; (2) existing data quality documentation does not adequately cover the diverse 2K error-types and report topics; (3) fast report-writing is critical for sailor relevance; and (4) sailors have diverse experience levels and needs. To address these challenges, RUBICON combines several innovative technologies and approaches: (1) the user interface converts LLM interactions into adaptive hints and user-prompts while avoiding slow chat-style workflows; (2) report quality rubrics (RQRs) encode requirements as natural language statements directly pulled from SME conversations; (3) adaptive retrieval systems uses structured and unstructured methods to standardize language and provide context to LLMs; and (4) separate information extraction and report generation steps reduce hallucinations in drafted reports. Currently, we are testing our prototype through live sailor demos for in-depth feedback. In this talk we will discuss the key features of RUBICON, describe the underlying technologies, and provide a demonstration of the user-interface.

Rubric-Based Intelligent Curation (RUBICON): Real-time sailor assistance for high-quality maintenance report writing

The integration of Artificial Intelligence (AI) into naval operations offers transformative capabilities for mission readiness, predictive analytics, and decision superiority. Yet, these benefits introduce governance challenges that demand a structured approach. AI governance must ensure transparency, compliance, and operational resilience while mitigating risks from unapproved or uncontrolled deployments.

A critical component of this framework is AI Discovery, which provides full visibility into all deployed models and tools across shipboard, shore-based, and cloud environments. Discovery enables inventory management, compliance checks, and identification of unauthorized AI usage, ensuring that only validated solutions are operational. This capability reduces risk and strengthens trust in mission-critical systems.

Governance should also include centralized policy enforcement, continuous monitoring, and risk scoring for every AI asset. Automated validation workflows should confirm adherence to ethical principles, cybersecurity standards, and mission objectives. Integrated dashboards should deliver real-time insights into model performance, bias detection, and anomaly alerts, while lifecycle controls manage versioning and retraining to maintain accuracy and reliability. Human-in-the-loop checkpoints remain essential for high-impact decisions, aligning with Department of Defense Responsible AI principles.

By institutionalizing these capabilities, the U.S. Navy can accelerate AI adoption while safeguarding mission integrity and national security. This approach ensures that innovation advances without compromising trust, legality, or operational control.

AI Governance at Sea

Government agencies are rapidly accelerating AI across mission-critical operations to protect citizen wellbeing and national security. This momentum brings significant risks in predictive, generative, and agentic systems, including bias, opacity, safety failures, and governance gaps that must be managed.

This session highlights how government agencies can adopt a unified AI Observability and Security platform as their command center for AI deployments, giving them the oversight and audit trails they need to deliver high-performance AI, safeguard public trust, and maximize mission effectiveness and accountability. It enables teams to operationalize scalable, performant, and safe applications and models through continuous monitoring, diagnostics, and interpretability. 

We will share a case study on Fiddler’s work with the Defense Innovation Unit (DIU) on Project AMMO (Accelerated MLOps for Maritime Operations). Fiddler AI contributed observability capabilities for post-deployment monitoring, image embedding-drift detection with interactive root-cause analysis/explainability that accelerated Automatic Target Recognition (ATR) model retraining for unmanned undersea vehicles (UUV). Following the successful prototype, Project AMMO transitioned to production with the Naval Information Warfare Center (NIWC) Pacific.


AI Observability and Security: Your Command Center for National Security

TorchSig is an open-source, radio frequency (RF) machine learning (ML) library with synthetic signal datasets, domain transforms, and pre-trained models. This briefing will cover the v2.1.0 release, new pre-trained models, and new open-source repositories under TorchSig.
TorchSig's datasets are synthetically generated, which allows for the robust creation of large RF datasets for ML. In the v2.1.0 release, we have added even more modularity to our dataset and signal generators. Users now have the most flexibility and control over the generated signals. Users can 1) parameterize the datasets, signal family, or even specific signal level, and 2) create their own signals. For instance, a signal can be parameterized to only generate in a certain bandwidth of frequencies. Or, a completely new signal (e.g., Morse code) can be developed and easily integrated into the TorchSig signal generation pipeline. In addition to our 50+ already supported signal classes, users can adapt the library to their specific applications—whether digital, analog, or completely different signals. Under the hood, this means a restructuring of dataset and signal metadata, as well as slight API changes.
TorchSig now also supports users importing their own collected data into the ecosystem, which we call Bring Your Own Data (BYOD). This requires the user to write their own file handler so the library knows how to interface with their data. This feature also allows the user to augment their data with TorchSig's transforms and impairments.
We have released two pre-trained models under the torchsig-models repository. The first model is for narrowband classification, trained on IQ data with a dataset size of 1.14 million. The second model performs wideband energy detection on spectrograms, with a dataset size of 385k. These models are a good starting point for users, and can be fine-tuned on application-specific data. Note that both models are trained on TorchSig v2.0.0.
Lastly, we have released two more open-source repositories that use TorchSig. One is called gr-spectrumdetect, which uses a pre-trained v0.6.1 TorchSig wideband model with GNU Radio for live RF energy detection. Additionally, we have agentic-spectrumdetect, which combines a high-performance TensorRT-accelerated YOLO model and an LLM agent for agentic spectrum analysis.
TorchSig is developed by researchers at the Laboratory for Telecommunication Sciences (LTS) in College Park, Maryland. The team actively pushes regular releases on GitHub. We are always looking for feedback on how to make our library better for as many needs and applications. For this purpose, either a talk or workshop session would be beneficial to discuss how TorchSig can support the Navy's missions.

More information about TorchSig:
TorchSig website: torchsig.com
TorchSig GitHub page: https://github.com/torchdsp/torchsig

Premium content

Downloads

Next from NAML 2026

Decision Engine for Situational Context, Autonomous Replanning, Task Execution & Scheduling

Similar lecture

Design, Demonstration, and Proof of Concept Using the Explainable Reinforcement Learning (XAI) of Soar for Combat Identification (CID)

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES