Keywords: RF Electronics, Enabling Technology for Artificial Intelligence
Abstract: Traditionally, chip-scale RF system design has been the domain of experts, dominated by thumb rules, trial-and-error techniques, standard templates, time-consuming parameter sweeps, and ad hoc population-based optimization. There is no reason to believe that this approach is optimal. This talk will discuss how inverse design with AI-based approaches can open new design spaces and enable rapid designs on demand. It will cover deep-learning-based modeling and generative-AI approaches that transfer across process technologies for inverse design and automated synthesis of mm-wave and sub-THz circuits and antennas.

Keywords: Digital Engineering, Robotics, Enabling Technology for Artificial Intelligence
Abstract: This paper investigates the reasoning capabilities of Large Language Models (LLMs) by adopting a problem-solving approach. For maritime situation awareness, ChatGPT-4 was tested on a practical task for ship identification by gradually integrating a variety of information sources and observations. Experiments were conducted by using data of variable types and quality and then measuring the consistency of results when several iterations are performed and similar prompts are used. The goal is also to test whether LLMs can support decision making and can be used in daily activities of security stakeholders. In contrast to accuracy analysis, where comparison to ground truth is available, this approach only covers comparisons on LLMs results in an effort to emphasize discrepancies. The experiments corroborate the fact that LLMs are not consistent in their responses, and their outcomes should be carefully examined before being considered for operational tasks. These findings offer deeper insights on LLMs reasoning capabilities and a new perspective for more nuanced utilization of LLM models for data fusion problem-solving tasks.

Keywords: Software and Networking
Abstract: We describe the design of a capture-the-flag like testbed and use it to study the cognitive vulnerabilities of human penetration testers and compare them against an LLM-based agent. We find significant differences between the humans and the LLM for most cognitive vulnerabilities. We conclude that testing the cognitive vulnerabilities of LLMs in cyber security settings require different testbeds than for humans due to differences in how LLM-based agents perceive and process their environment, and these differences influence how defensive measures targeted at LLM-based should be implemented.

Keywords: Software and Networking, Digital Engineering, Trusted and Assured Hardware
Abstract: This work presents an AI-assisted digital control firmware for lithium-ion battery chargers operating under drift, stress, and aging. A lightweight RLS estimator with adaptive forgetting updates plant parameters for real-time gain adjustment, while safety-aligned fault management ensures deterministic recovery. Hardware-in-the-loop validation demonstrates improved stability, efficiency, and fault resilience.

Keywords: Trusted and Assured Hardware, Digital Engineering, Software and Networking
Abstract: This study evaluates Correlation Power Analysis (CPA) attacks on AES-128 ECB implementations across embed- ded platforms, with primary experiments on the ESP32-S3 and comparisons to ESP32-H2 and Arduino Uno R3(ATmega328P). A repeatable measurement setup using a USB oscilloscope and FPGA based acquisition system was developed to capture power traces. The signals were preprocessed and analyzed using Pearson correlation with a Hamming-weight model. The results show that key recovery is feasible under controlled conditions, but increased architectural complexity introduces switching noise that reduces leakage visibility. Hardware modifications, such as amplification and capacitor removal, improved signal quality, emphasizing the importance of measurement design in side-channel analysis.

Keywords: Autonomous Guidance, Navigation and Control, AI for Cognitive Sensing and Data Fusion, Enabling Technology for Artificial Intelligence
Abstract: This talk presents a multi-scale framework for scalable robotic autonomy, modeling large robot teams as evolving distributions while enabling decentralized agent-level adaptation. The framework connects distribution-based planning, distributed control, and mobile sensing networks to support resilient coordination under uncertainty, limited communication, and real-world deployment constraints.

Keywords: Enabling Technology for Artificial Intelligence, Software and Networking, Communication
Abstract: Modulation classification classifies modulation schemes on RF signals without prior coordination between a transmitter and a receiver. Although current lightweight neural networks perform well on standard simulated datasets, their robustness given domain shifts as well as in real-world environments remains unverified. This study investigates the robustness of lightweight neural networks running on FPGAs when exposed to severe domain shifts in RF signals. We produce simulated RF signals with various channel fading and hardware imperfections and collect real-world I/Q frames from USRPs. We generate hardware implementations of quantized neural networks with the FINN framework and perform cross-domain testing with these quantized neural networks running on FPGAs. Our findings demonstrate that (1) domain shifts between training RF data and test RF data indeed cause significant accuracy degradation; (2) discrepancy in hardware imperfections can lead to more severe accuracy degradation compared to channel fading discrepancy; (3) accuracy degradation is not even across modulation types, where predictions on a subset of modulations may still perform reasonably well; (4) the accuracy reported in FPGAs is not consistent with the one observed in GPUs, highlighting the importance of reporting results in FPGAs rather than using GPUs only for RF classifications.

Keywords: AI for Cognitive Sensing and Data Fusion, Communication, Terrestrial and Space Applications.
Abstract: The deployment of multi-view three-Dimensional (3D) reconstruction in distributed environments incurs substantial communication overhead. This paper proposes a multi-view 3D reconstruction framework based on fine-grained region selection that reduces bandwidth demand while preserving reconstruction-critical information, enabling accurate 3D reconstruction with lower communication cost than representative view-selection methods in edge-to-server settings.

Keywords: Robotics, Communication, Terrestrial and Space Applications.
Abstract: This keynote presents an overview of the evolution of space robotics, from the first orbital manipulator to today's intelligent robotic systems for on-orbit servicing, in-space manufacturing, and exploration. It reviews recent advances in sensing, manipulation, artificial intelligence, and human–robot collaboration, and discusses the technological challenges that remain. Finally, the talk offers a vision for the future of space robotics and its emerging directions.

Keywords: Robotics, Enabling Technology for Artificial Intelligence, Hardware-in-the-Loop (HIL) Testing
Abstract: Robotic learning can revolutionize labor in unstructured environments like homes. However, most robot manipulators are expensive or position-controlled, limiting deployability. This paper presents the design and Mixed Reality control of the torque-controlled TinyARM5 robot, a safe, inexpensive (<300 USD), and compliant desktop manipulator. Open-source release: www.nicks-robots.com/tinyarm5/.

Keywords: Architecture, Terrestrial and Space Applications., UAVs
Abstract: Electroaerodynamic (EAD) "ion" propulsion offers a silent and combustion-free alternative for ultralight aircraft, but has historically been constrained by the mass of thrusters powerful enough for practical autonomous aerial systems. This paper presents an optimization of electrode geometry via a sin(θ/2) plateau model combined with a novel ultralight balsa-PLA hybrid construction method, achieving an 87.5% structural mass reduction to 6g while preserving electrode alignment. The optimized thruster was integrated into a 15 g blended-wing airframe and validated through indoor flight testing, with a 431% mean distance improvement and 60% mean time-aloft improvement over unpowered glides. To address performance under realistic conditions, the thruster was validated using a compressor-driven wind tunnel rig with load-cell measurement under a 3 m/s headwind and 40° crosswind, confirming sustained thrust generation under both flow conditions. These results are a scalable pathway toward ultralight EAD propulsion for humanitarian and low-noise urban aerial applications.

Keywords: Software and Networking, Trusted and Assured Hardware
Abstract: This presentation explores the impact of Artificial Intelligence on the "cyber problem" from a U.S. Air Force perspective. It details the evolving threat landscape, where AI enhances both adversarial capabilities and defensive cyber operations. The brief covers current research, doctrinal guidance, and major gaps to be addressed for future warfare.

Keywords: Software and Networking
Abstract: In this paper, we model and analyze an attacker’s decision-making process during reconnaissance using a hypothesis-testing framework. To deceive the attacker during reconnaissance, the defender is considered to strategically and heterogeneously falsify information observed by the attacker. Performance limits of deception and optimal budget-constrained falsification strategies are characterized, and numerical results providing key insights are presented.

Keywords: Trusted and Assured Hardware, Software and Networking
Abstract: With the increase of wireless transmitting devices, identification of emitters becomes necessary. Radio frequency (RF) fingerprinting is an active field of study that focuses on identifying unique features of a transmitted signal, that is, ``fingerprints'' that capture qualities distinct to the given transmitting hardware. An important factor to consider is the impact of the domain in which a signal is transmitted. We focus on the problem of receiver variance and use a collaborative-adversarial technique on the open-access dataset, WiSig. We conduct a comparative study of domain generalization techniques across different deep learning architectures, and we verify performance improvement over an equivalent model without the additional training mechanism. Furthermore, we show how this technique impacts the saliency of the models and generalizes the features to recognize correct classes across receivers.

Keywords: Enabling Technology for Artificial Intelligence, AI for Cognitive Sensing and Data Fusion, Software and Networking
Abstract: Large language models (LLMs) can rapidly generate synthetic social-media content, increasing the need for practical detection methods. Because general-purpose LLM-text detection remains difficult, this study evaluates a narrow-domain approach using simple numeric text features and classical machine-learning models. On Twitter/X data, the QDA classifier selected through training-set cross-validation achieved 0.9608 accuracy and a 0.9590 F1 score in distinguishing human-authored tweets from GPT-3.5-generated tweets. These results demonstrate the potential of simplified feature-based detection within constrained domains.

Keywords: Enabling Technology for Artificial Intelligence
Abstract: Large Language Models (LLMs) are useful tools for information retrieval, but they do not remember a user’s preferences or interaction history across various sessions. Thus, we propose a novel knowledge graph based memory system that works with any tool capable LLM to return a user’s preferences based on their past conversations. This makes it easier for the LLM to personalize how it responds to the user’s query, and how it searches for information on the user’s behalf. Our system uses a single knowledge graph per user for storing user interaction history and preferences, is LLM agnostic, and is adapted for personalized organizational data search.

Keywords: Enabling Technology for Artificial Intelligence, Digital Engineering, Trusted and Assured Hardware
Abstract: Designing cyber-physical systems (CPS) that minimize vulnerabilities is difficult even with security engineering expertise. As a system becomes increasingly complex in terms of size, functionality, or usability, security analysis quickly becomes stale. Reasoning about vulnerability mitigation could be assisted with large language model (LLM) technology. This paper presents a methodology that progressively qualifies LLMs on the task of analyzing systems for attack paths and countermeasures, using Bloom’s Revised Taxonomy as a task-language refinement instrument: each successive rung is treated as a behavioral target to qualify prompt language. The methodology is applied to two open-weight 7–8 billion parameter LLMs deployable on 8GB VRAM hardware. The primary result is diagnostic: at the Understand rung the models default to an enterprise information technology (IT) framing of security terminology, requiring the prompt to carry the cyber-physical context explicitly rather than leaving it to the model’s training. Comparisons of the base task versus reasoning scaffolds such as Chain-of-Thought did not detect an improvement at this parameter scale; a structural analysis of the outputs offers a candidate explanation: models often emitted reasoning after the answer, where it cannot condition the result under autoregressive decoding. Future work may re-use the methodology on LLMs with greater parameter counts.

Keywords: Robotics, Terrestrial and Space Applications.
Abstract: As space missions grow in complexity, the importance of in-space servicing, assembly, and manufacturing is increasing. Space manipulator systems are essential for these operations, yet the dynamic coupling effects in free-floating environments can hinder the system's stability. This work presents the construction and experimental validation of a reconfigurable dual-arm-mounted free-floating spacecraft testbed. Unlike existing testbeds with fixed parameters, the proposed system allows for systematic adjustment of link lengths and mass properties to study their influence on the dynamic coupling. Experimental results demonstrate reliable tracking performance of joint controllers. Furthermore, the implementation of reaction-minimized manipulation successfully reduces the base's reactive motion by approximately five times compared to standard operation. This work establishes a multi-purpose on-ground testbed for verifying diverse in-space servicing mission scenarios.

Keywords: Communication, Networking, Terrestrial and Space Applications.
Abstract: This paper presents OLEC, an Online-Learning-Based error concealment method to improve the robustness of point cloud streaming. It detects corrupted regions due to packet losses and reconstructs them via lightweight spatial-temporal regression. Evaluation results demonstrate that OLEC improves the perceptual quality of received point clouds compared to benchmarks.

Keywords: Autonomous Guidance, Navigation and Control, bio-inspired sensors, Robotics
Abstract: This paper presents a gaze-based human–robot interaction (HRI) framework for intuitive control of a robotic manipulator. The proposed system utilizes an eye tracking sensor to continuously capture user gaze behavior and infer interaction intent for guiding robot actions in collaborative and assistive tasks. Gaze information is processed and mapped to robot motion commands, enabling natural and non-invasive control without reliance on traditional programming interfaces. The framework is designed to support applications in assistive technologies, teleoperation, and rehabilitation robotics, where users may have limited motor capabilities. We use gaze-to-action mapping and intent disambiguation to improve robustness and reliability under real-time conditions. Experimental evaluations demonstrate that users can effectively guide robotic manipulation and pointing, directing-to/steering tasks using gaze alone, confirming the feasibility of the proposed approach. The results highlight the potential of gaze-guided control to reduce user effort, enhance accessibility, and improve the efficiency of human–robot collaboration compared to conventional control methods. The proposed gaze-based control framework is validated through the successful execution of an insertion-like task, demonstrating its capability for fine and precise robot manipulation with a positioning accuracy on the order of 75 μm from the desired target.

Keywords: Microelectronics
Abstract: Semiconductors and Microelectronics provide the differentiated backbone required to deliver and enhance national and economic security. As the complexity of microelectronics technology increases from materials to test to integration, the cost and availability of access to prototyping and production capability becomes prohibitive to a robust and resilient innovation pipeline and supply chain. This talk will focus on MMECs approach and solutions to enable a next-generation acquisition and sustainment pathway for advanced electronics capability through partnership with academia, commercial technology companies, traditional and non-traditional defense contractors, manufacturers, and system integrators. The presentation will outline the motivations, origins, capabilities, and roadmap of MMECs CONNECT Platform -- providing access to a secure and scaled ecosystem with infrastructure, expertise, tools, IP, and production systems that demonstrate measurable technical advantage for DoW and other national-security applications.

Keywords: AESAs, Electronic Warfare and Countermeasures, Communication
Abstract: Fully digital RF sensors promise unprecedented sensing capability but create extreme processing demands. This keynote examines scalable on-array processing architectures developed in DARPA’s Scalable On-Array Processing (SOAP) program. Leveraging Intel’s Configurable Spatial Accelerator, distributed processing enables autonomous and AI-driven RF sensing while meeting stringent size, weight, and power constraints of future platforms.

Keywords: Trusted and Assured Hardware, Enabling Technology for Artificial Intelligence, Collaborative/Multi-Agent Systems
Abstract: Materials which are used in the manufacturing of microelectronic components are vulnerable to malicious contamination of alloys, isotopes, chemical kinetics and crystalline defects in the supply chain. We propose an agentic AI system which identifies material vulnerabilities. Example prompts are provided for identifying the lifetime failure rate crossover for components.

Keywords: Software and Networking
Abstract: This study concentrates on the efficacy of ChatGPT in penetration testing. We measured performance of penetration testing on 5 virtual machines with Kali using 2 metrics, the time taken to compromise each machine and a Likert scale. ChatGPT use did speed up attacks, but efficacy dropped when requesting specialized knowledge, indicating GPT-5 could be a hindrance.

Keywords: Trusted and Assured Hardware
Abstract: RowHammer exploits repeated DRAM row activations to corrupt adjacent memory cells. Hammulator, a gem5-based simulator, enables rapid prototyping—but its cross-generational reliability is undocumented. Our experiments reveal stark asymmetry. DDR4 scales predictably: bit flips grow from 12 to 1,452 as hammer iterations increase from 10,000 to 50,000. DDR3, under identical conditions, produces zero bit flips—no errors, no warnings. The cause: Hammulator's default DDR3 configuration lacks fault-model parameters present in DDR4. Manually inserting them still yielded no DDR3 bit flips, implying undocumented integration dependencies. This contradicts hardware studies showing 85% of DDR3 chips are physically vulnerable. Using Ramulator2, we confirmed both configurations generate sufficient memory pressure—isolating the failure to Hammulator's fault-injection layer. Researchers running cross-generational studies risk drawing false conclusions about DDR3 resilience. Simulation frameworks must provide verified multi-generational support or explicitly document limitations. We recommend configuration validation mechanisms to prevent silent failures from undermining DRAM security research.

Keywords: Autonomous Guidance, Navigation and Control, Enabling Technology for Artificial Intelligence, Robotics
Abstract: Autonomous systems that build structures from natural-language instructions need reliable spatial reasoning, yet large language models (LLMs) make systematic coordinate errors when generating three-dimensional block placements. We present a neuro-symbolic pipeline based on 2.5-D decomposition: the LLM plans in the two-dimensional horizontal plane, while a deterministic executor computes all vertical placements from column occupancy, eliminating an entire class of errors. On the Build What I Mean benchmark (160 rounds), GPT-4o-mini with this pipeline achieves 94.6% mean structural accuracy across 12 independent runs, within 3.0 percentage points of the 97.6% ceiling imposed by architect-agent errors that no builder-side improvement can address. This outperforms both GPT-4o at 90.3% and the best competing system at 76.3%. A controlled ablation confirms that 2.5-D decomposition is the dominant contributor, accounting for 28.7 percentage points of accuracy. The pipeline transfers directly to edge hardware as shown in experiments with the Nemotron-3 120B on an NVIDIA Jetson Thor AGX. The underlying principle, removing deterministic dimensions from the LLM's output space, applies to any autonomous construction or assembly task where gravity or other physical constraints fix one or more degrees of freedom. A transfer experiment on 500 IGLU collaborative building tasks confirms the effect generalizes beyond the primary benchmark.

Keywords: Collaborative/Multi-Agent Systems, Terrestrial and Space Applications., Simulation-Based Validation and Verification
Abstract: Reinforcement learning (RL) policies for Low Earth Orbit (LEO) satellite constellation scheduling are typically trained in simplified simulations for computational tractability, yet must operate under realistic orbital dynamics. This pa per investigates how training-environment fidelity affects policy transfer to the high-fidelity Basilisk astrodynamics simulation. We evaluate two hierarchical multi-agent architectures: (1) a Centralized Training, Decentralized Execution (CTDE) actor critic policy trained with Keplerian two-body dynamics, and (2) a gradient-based Vertical Federated Reinforcement Learning (VFRL) policy trained in a flat two-dimensional grid with no orbital model. Both are deployed in an identical Basilisk Walker delta simulation via a flatten_norm mapping function that bridges three-dimensional orbital state to two-dimensional grid observations. Over 50 evaluation episodes, the Kepler-trained policy achieves a mean task completion rate of 79.0% ± 2.7%, outperforming the 2D-trained policy at 70.3% ± 6.1% (Welch’s t test, p < 10−9). These results demonstrate that even approximate orbital dynamics during training provide substantial transfer benefits, and that mapping functions are necessary but not sufficient: transfer quality depends critically on source-domain fidelity.

Keywords: AI for Cognitive Sensing and Data Fusion
Abstract: Cross-camera human identity tracking remains challenging in intelligent surveillance systems due to viewpoint variation, occlusion, and fragmented trajectories across distributed camera networks. Existing approaches either sacrifice real-time performance through computationally intensive transformer-based optimization or rely on large-scale identity galleries that become increasingly costly as tracked identities accumulate. This paper presents VigiTrace, a lightweight multimodal framework for real-time cross-camera identity tracking using an Adaptive Resonance Theory (ART)-based association mechanism that jointly leverages appearance, pose, motion, and temporal consistency cues within compact adaptive identity prototypes. Unlike gallery-dependent methods, the proposed approach supports incremental identity learning and open-set identity management without exhaustive feature matching. Evaluated on the Market-1501 benchmark, VigiTrace achieves an mAP of 87.6 while maintaining real-time operation at 30 FPS. Experimental results demonstrate competitive identity association performance with substantially higher runtime efficiency than transformer-based baselines operating at significantly lower frame rates.

Keywords: Extreme Environment Electronics
Abstract: Resonant Sciences is a leader in electromagnetics and has delivered tools and hardware that solve some of the most challenging problems across academia, industry and government. These solutions guarantee performance, reliability, and safety in the world’s harshest environments. During this presentation, our three business areas will be presented with specific examples. Vertically integrated, our company brings new technology from theory to reality, ensuring we maintain quality control throughout our product development. Standing on our success, our company is growing rapidly and we are seeking qualified individuals across engineering disciplines to support our growth. I hope you can find time to listen to my talk or find me at the Resonant Sciences booth.

Keywords: Enabling Technology for Artificial Intelligence, AI for Cognitive Sensing and Data Fusion, Autonomous Guidance, Navigation and Control
Abstract: With the rapidly evolving organization functioning, digital integration, and wide-ranging cross collaborations, it is imperative to have an efficient, and dedicated tool for expeditious generation of engineering deliverables such as documentation, and performance metrics of innovative products. Most of such existing tools suffer from limitations such as manual interventions, fragmented documentation, reduced formats and types of documents, and higher time consumption. In this paper, we introduce an end-to-end Generative AI driven engineering deliverables generation chatbot that is capable of performing tasks including but not limited to generating, updating, and rectifying multi-format engineering deliverables. This chatbot developed in collaboration with Texas Instruments for expediting their internal operations was validated using open-source materials. Evaluation results demonstrates the feasibility of our chatbot besides providing the resource requirements in terms of memory, and computational time in completing basic tasks.

Keywords: Autonomous Guidance, Navigation and Control, Collaborative/Multi-Agent Systems, Terrestrial and Space Applications.
Abstract: We address the problem of multi-spacecraft formation acquisition by comparing a traditional LQR+APF controller with a reinforcement learning based agent. The scenario demonstrates maneuvering three deputy satellites under discretized Hill–Clohessy–Wiltshire dynamics. The operational environment imposes hard keep-out and collision-avoidance constraints. For the guidance task, we employ a multi-agent PPO (MAPPO) controller. The training pipeline leverages centralized training/decentralized-execution and is initially warm-started through behavior cloning. In a paired 200-episode hold-out evaluation, the final learned controller achieves 0.805 success versus 0.850 for the LQR+APF baseline. The paired comparison does not detect a statistically significant difference (p = 0.175), and neither controller incurs a hard safety violation. The results validate the learning pipeline while demonstrating a reinforcement learning based controller can perform similarly to an LQR and maintain safety regions.

Keywords: Enabling Technology for Artificial Intelligence
Abstract: Abstract— Conventional intuition suggests more training data improves neural network robustness. A factorial MNIST study shows this advantage under clean inputs collapses under data degradation, and can invert under occlusion. Results suggest data scaling alone does not ensure robustness to degraded sensing conditions.

Keywords: Autonomous Guidance, Navigation and Control, AI for Cognitive Sensing and Data Fusion, Enabling Technology for Artificial Intelligence
Abstract: Inferring flight intent from partial Automatic Dependent Surveillance–Broadcast (ADS-B) trajectories is a critical challenge for aviation analytics, particularly under incomplete or degraded observations. This work presents an LLM-assisted framework for joint airport classification and trajectory completion using partial flight segments. Trajectories are represented as time-series data and normalized via cubic spline interpolation, while stochastic segmentation simulates real-world partial observability.

We evaluate state-of-the-art classification models, including Long Short-Term Memory (LSTM), InceptionTime (IT), and MiniROCKET (MR), alongside an encoder–decoder LSTM for trajectory completion. Experiments across multiple segment lengths and random seeds demonstrate that partial trajectories contain sufficient information for accurate classification, with IT performing best under limited observations and MR benefiting from longer inputs. The seq2seq model achieves stable prediction performance, though accuracy degrades under route ambiguity in dense airspace.

A human-in-the-loop workflow using GenAI.mil accelerates development while preserving domain validity. Results highlight the feasibility of joint classification and prediction and demonstrate the value of LLM-assisted methodologies for data-driven aviation research.

Keywords: Enabling Technology for Artificial Intelligence, Digital Engineering, Neuromorphic Computing
Abstract: This paper presents a tiled systolic MAC array for FPGA-based neural network acceleration in two stages. Stage-I stores input activations inside the MAC units to achieve hidden layer reuse and weight refeeding. Stage-II implements the idea in RTL on a Xilinx PYNQ-Z2 platform and tiled execution. Cycle-accurate simulation gives quadratic scaling per layer and reaches 97% MAC utilization. FPGA implementation on Vivado toolset verify operand alignment, scheduling and forwarding in both stages.

Keywords: Enabling Technology for Artificial Intelligence, Collaborative/Multi-Agent Systems, Digital Engineering
Abstract: Generative AI (GenAI) tools, specifically LLMs, have seen an explosion in use across professional fields in recent years. Researchers have explored integrating GenAI tools into Multi Agent Systems (MAS) for automated engineering design pipelines. However, engineering subfields contain domain knowledge which includes tacit expertise that is often missing from LLM training data. Applying LLMs without tacit knowledge to domain-specific tasks increases the risk of injecting false or misguided information into the design process. These risks are even further amplified when using autonomous Agentic LLMs that can act independently. Knowledge Graphs (KGs) can improve LLM behavior, but manual domain KG creation is resource intensive. Automating KG creation via Knowledge Graph Extraction (KGE) using LLMs is becoming more common, but its implementation is underexplored in engineering design pipelines. This paper proposes a meta model to explore the tradespace behind LLM-based options for MAS engineering design workflows.

Keywords: Software and Networking, Enabling Technology for Artificial Intelligence, Digital Engineering
Abstract: Behavioral analysis has proven effective for ana- lyzing software yet remains largely unexplored for embedded firmware. Traditional vulnerability detection techniques struggle on Linux-based firmware due to heterogeneous file systems, diverse processor architectures, and the absence of source code. Critically, many embedded vulnerabilities are emergent: they arise not from any single dangerous file but from a chain of interactions across the file system. We propose representing a firmware image as an annotated, heterogeneous directed graph in which nodes correspond to files and edges correspond to inter- file call relationships, and training a Graph Neural Network (GNN) on these graphs to perform multi-label vulnerability classification. Each node carries a bill of behaviors—a feature vector of security-relevant capabilities extracted by a rules- based engine—and each edge encodes call semantics such as user-controlled arguments and privilege-boundary crossings. This paper presents the methodology, feature schema, and a rigorously specified evaluation design. As this is a pre-implementation proposal, we report no results; instead we define the metrics, baselines, and leakage-free splitting protocol against which the approach should be measured, and we identify ground-truth labeling as the central feasibility risk.

Keywords: Neuromorphic Computing, AI for Cognitive Sensing and Data Fusion, Enabling Technology for Artificial Intelligence
Abstract: Real-time ISR demands sub-millisecond multi-modal classification under tight SWaP, SLA and provenance constraints. A 46-node BrainChip Akida fleet under IBM Spectrum Symphony over GPFS fuses LPI radar, photonic RF, neuromorphic vision and OSINT modalities, delivering sub-200 ms autonomous kill-chain decisions and a 30-watt 120-chip helmet-distributed perceptual layer. A demonstration program of sixty-seven builds exercises the arrangement across kinetic ISR, small-unit perception, orbital sensing and civilian remote sensing, and the composition results of Grindrod, van der Made and Motta and Nanni supply the dynamical, anatomical and empirical warrant for federating modest neuromorphic units rather than enlarging a single one.

Index terms: neuromorphic computing, BrainChip Akida, ISR, multi-sensor fusion, LPI radar, photonic RF, OSINT, SWaP, IBM Spectrum Symphony, GPFS, CKKS, compute ontology, federated neuromorphic learning

Keywords: Enabling Technology for Artificial Intelligence
Abstract: Proprietary constraints on real-world flight data have driven aerospace AI researchers toward X-Plane-based synthetic data generation; however, each study independently reimplements collection pipelines, impeding reproducibility and cross-study comparison. This paper presents FlightForge, a four-layer framework comprising a declarative YAML scenario specification language and a Python orchestration layer built atop NASA's X-Plane Connect (XPC), producing standardized, labeled HDF5 datasets. FlightForge is validated across two representative use cases: guidance, navigation, and control; and anomaly detection.

Keywords: Enabling Technology for Artificial Intelligence, Software and Networking, Collaborative/Multi-Agent Systems
Abstract: Electronics design automation (EDA) tools play a critical role in microelectronics design of product development through integrated circuits. We propose an agentic AI system which autonomously reasons, plans and initiates responses to EDA infrastructure management and risks. Example prompts are provided for license recommendations for a software license agreement.

Keywords: Digital Twins, Microelectronics, AI for Cognitive Sensing and Data Fusion
Abstract: Mixed-signal integrated circuits deployed in long life-cycle applications such as defense, aerospace, and space systems must remain operational for decades after initial deployment, even as the original technology node, design tools, and engineering resources become obsolete. This paper presents a digital twinning methodology that enables reproducible regeneration of mixed-signal systems across technology nodes and design environments without redesigning the functional architecture. The proposed framework preserves the complete implementation context—including the register-transfer-level (RTL) description, configuration parameters, and verification environment—allowing systems to be reconstructed when fabrication processes or toolchains are no longer supported. The methodology is demonstrated using a clock and data recovery (CDR) subsystem employing an analog time-encoding front end based on a voltage-controlled oscillator (VCO) and digital machine learning filter architecture, where technology-dependent behavior is localized to delay elements while digital logic remains technology-independent. Experimental validation across multiple CMOS technology nodes using an automated RTL-to-GDS implementation flow demonstrates consistent functional behavior and scalable implementation, establishing digital twinning combined with analog time encoding as a practical strategy for sustaining mixed-signal systems in mission-critical applications over multi-decade lifetimes.

Keywords: Robotics, Digital Twins, Enabling Technology for Artificial Intelligence
Abstract: This paper presents a virtual-reality pipeline for robot demonstrations with a custom 5-DoF arm in Isaac Sim and Isaac Lab. Demonstrations are replayed, retargeted to joint-space actions, and re-rendered with new sensors. The workflow collected 200 demonstrations in 30-minute trials compared with 45 real-world demonstrations and supported behavior-cloning policy evaluation.

Keywords: Quantum Computing, Quantum Sensing and Timing, Quantum Networking
Abstract: With the exponentially increasing global data volumes it is challenging to meet the ultra-low latency requirements of the next-generation sophisticated applications using the existing computing resources. Though multiple existing methods and approaches such as edge-and-cloud computing, high performance computing (HPC), and specialized accelerators address the processing of voluminous data they suffer from numerous drawbacks. Some of them includes higher power consumption, expensive installation process, and lack of data privacy mechanisms. Quantum computing due to its inherent properties such as entanglement, and superposition becomes an ideal candidate for processing high-dimensional as well as exponentially growing data requirements. However, majority of the existing quantum computing researches focus on the individualized circuit implementation. But the modular quantum testbed is integral to systematic circuit analysis within emerging quantum computing architectures. In this paper, we propose Python based multi-layered quantum testbed that integrates modular circuit construction, simulation, and measurement analysis. Additionally, this paper includes preliminary arithmetic results from our testbed besides including a comparison between classical and quantum computational algorithms. This testbed will lay a solid foundation towards building a sophisticated quantum architectures for security related applications.

Keywords: Autonomous Guidance, Navigation and Control, AI for Cognitive Sensing and Data Fusion, Collaborative/Multi-Agent Systems
Abstract: Synthetic image generation is integral to the testing, and validation of highly sophisticated image projection systems such as displays, and other pertinent applications involving image analysis. Furthermore expeditious testing of innovative projection system is pertinent for the dole out of next-generation sophisticated display based products. However, most of the existing works in synthetic image generation suffer from limitations such as the ability to generate one specific application image, inability to standardize the generated synthetic images across diverse testing environments, and inability to replicate. To address these limitations, we propose the development of user-friendly AI-powered multi-layer interactive platform that can generate diverse user-specific images with desired configurations. Salient aspects of our platform includes flexibility in addition or deletion of features, scalability, adaptability, security, and cost efficient. Preliminary evaluation results underscores the feasibility, and existence of our proposed platform.

Keywords: Autonomous Guidance, Navigation and Control, Robotics, UAVs
Abstract: This work presents a rate-aware optimal control framework for fixed-wing unmanned aircraft systems that incorporates actuator rate dynamics and outer-loop tracking compensation into the optimization. Simulations in turbulent wind conditions show the controllers maintain strong tracking performance while reducing aircraft control rates compared to standard optimal control methods.

Keywords: Neuromorphic Computing, Enabling Technology for Artificial Intelligence, AI for Cognitive Sensing and Data Fusion
Abstract: Modern large-language-model infrastructure routes nearly every request through dense GPU compute regardless of reasoning depth required. A spiking neural network workload manager on a BrainChip Akida cluster under IBM Spectrum Symphony's compute ontology routes across three tiers, dispatches in 5 ms and reduces GPU cost 30 to 50 percent. Beyond routing, the neuromorphic tier answers the same vLLM endpoints a GPU answers, holds 8,832 concurrent inference contexts across a 46-node fleet at 99.4 percent of ideal scaling and hot-swaps a hundred resident models at 11 ms each, which places the composition results of Grindrod, van der Made and Motta and Nanni on an operational substrate.

Index terms: neuromorphic computing, BrainChip Akida, IBM Spectrum Symphony, GPFS, compute ontology, LLM serving, vLLM, KV cache, semantic routing, sustainable AI, federated neuromorphic learning

Keywords: Autonomous Guidance, Navigation and Control, Robotics, Terrestrial and Space Applications.
Abstract: This paper explores and evaluates both traditional and learning-based methods for spacecraft control and coupled robotic arm manipulation in a microgravity environment. Space junk, debris, and out-of-control satellites currently floating in orbit pose a severe risk to critical space assets, necessitating debris removal systems. Controlling a robotic arm and base vehicle in zero-gravity environments poses a significant challenge due to the inherent physical coupling of the system's dynamics. This paper is concerned with a 3-DOF planar spacecraft with a 2-DOF manipulator approaching and intercepting a rotating object in 2D space. It establishes a cascaded Proportional-Integral-Derivative (PID) controller as a baseline benchmark before focusing on comparing optimal and learning-based control strategies, specifically a unified 5-Degree-of-Freedom Non-linear Model Predictive Control (MPC) and Deep Reinforcement Learning (DRL). The analysis evaluates steady-state precision, dynamic stability, and approach trajectories to determine the operational suitability of each methodology for target interception and maintained contact.

Keywords: Digital Twins, Simulation-Based Validation and Verification, Digital Engineering
Abstract: Mission-critical facilities generate heterogeneous power-event evidence from UPS controllers, EPMS platforms, power-quality meters, ATS/generator controllers, protective relays, switchgear, and BMS systems. Existing monitoring tools provide event capture, alarm grouping, waveform viewing, and operational visibility, but facility-level RCA still requires cross-domain interpretation across vendor-specific semantics, unequal timestamp quality, incomplete observability, and topology-constrained causality. This paper presents GridScope, a deterministic causal interpretation layer that reconstructs physically plausible root-cause hypotheses from existing monitoring evidence. GridScope combines vendor-neutral event normalization, RMS-level COMTRADE/PQDIF ingestion, topology-constrained causal filtering, uncertainty-aware timing, evidence-quality weighting, common-mode detection, protection-selectivity rules, external-context gating, and tri-state outputs. On a generated 90-scenario benchmark, GridScope achieved 100.0% top-1 accuracy under observable conditions and 95.3% selective accuracy overall after ratio-based ambiguity detection. Threshold sweeps place the selected ambiguity boundary on a broad performance plateau. Supplemental checks using real COMTRADE records, seven reconstructed documented incidents, and external-context datasets evaluate ingestion correctness and conservative decision behavior. Proprietary live facility-log validation remains future work.

Keywords: Enabling Technology for Artificial Intelligence, Semiconductor Devices
Abstract: 5:30-5:40pm Welcome: Dr. Pamela Abshire, Chair and Professor, Stony Brook University 5:40-6:05 pm Panel Introduction and Position Points: 5 minutes position points per panelist.

Panel Members: Intel: TBD IBM: Dr. Martin Frank Ozark Integrated Circuit: Dr. Matt Francis Siemens: Dr. Antonio de la Serna AFRL: Dr. Christopher Reuter University: TBD 6:05-6:35 pm Q&A and Panel Discussions (Moderator: Dr. Pamela Abshire, Chair and Professor, Stony Brook University) 6:30-7:15 PM Food, Drinks, and Networking