Previous Work and Case Studies

The U.S. Navy requires advanced, real-time tools to visualize and assess complex shipboard environments using large-scale 3D data from LiDAR and photogrammetry. Current workflows for system installation, maintenance planning, and configuration control are hindered by the lack of interactive, collaborative, and intelligent systems that can support distributed engineering teams. There is a mission-critical need for a solution that combines physics-aware rendering, anomaly detection, and multi-user collaboration in a secure, adaptive, and intuitive environment to reduce integration risks, identify spatial conflicts, and accelerate decision-making during ship modification and sustainment efforts.

To address this, CSLabs is conducting a Phase I feasibility study under Proposal N251-033-1949 to develop a mixed-reality (MR) system that blends physical and digital worlds for immersive, real-time interaction. The system will ingest and preprocess 3D models derived from LiDAR and photogrammetry, applying AI-driven segmentation, alignment, and anomaly detection to compare against reference data. A GPU-accelerated rendering engine will provide adaptive level-of-detail control, ensuring high performance across large, complex models. CSLabs’ MR workspace will also support real-time, multi-user collaboration, including spatial annotations, cloud version control, and automated clash detection—creating a shared, synchronized environment for shipboard integration planning and engineering analysis. This solution will ultimately enhance coordination, improve accuracy, and streamline complex shipboard workflows across the Navy enterprise.

The Navy is seeking to enhance its ability to execute complex maritime missions using UMVs across both surface and subsurface domains. These missions include unmanned transport of heavy payloads, surveillance, force protection, search and rescue, blockading, mine countermeasures, and Anti-Submarine Warfare (ASW). Next-generation UMVs with swarm capabilities are needed to complement manned platforms and expand operational reach. Autonomy that enables predictive behaviors and collaboration among diverse UMVs will improve mission effectiveness in highly dynamic and uncertain environments.

This technology is an autonomy framework that enables a UMV swarm to coordinate and adapt their actions based on mission objectives, environmental context, and shared situational awareness. The system supports behaviors such as patrolling, blockading, pursuing, and intercepting. Each platform adjusts its behavior dynamically, enabling effective multi-vehicle collaboration with minimal operator input and improving mission performance across a range of operational scenarios. In-water testing will be conducted to demonstrate core behaviors, with additional testing planned to support integration and transition. When technology hits technology readiness level (TRL) 6 it will enable collaborative autonomous UMV teams to support a wide range of Navy missions while reducing the need to place personnel in harm's way. Swarm-based behaviors such as carrying heavy payloads, search and rescue, surveillance, and blockading can be conducted without direct operator control, increasing responsiveness and reducing risk.

There is a critical need to modernize explosive breaching capabilities to address capability gaps created by rapidly advancing adversarial technologies and aging legacy breaching equipment. Current methods lack the speed, precision, and adaptability required to neutralize obstacles in complex minefields. A solution is required that leverages Artificial Intelligence and Machine Learning to rapidly detect and classify obstacle locations, generate optimized firing plans, and enable safe, efficient navigation routes for vehicles and personnel through contested terrain.

CSLabs developed AI-Routes, it is a framework and extension to DeltaX that automates route planning through minefields and creates firing plans to enable optimal routing using AI models for both minefield prediction and route planning. AI-Routes, revolutionized minefield path and Explosive Breacher (EB) firing plans, closing the capability gap between the asymmetric and quickly evolving adversarial minefield capabilities that have surpassed legacy mine breaching systems. The manual process of path planning and EB breach firing plans is inefficient and slow. AI-Routes automates route planning through minefields and creates firing plans to enable optimal routing, reducing overall mission risk. AI-Routes ingests minefield data and rulesets, sensor detection data, and additional geographic information system (GIS) data, utilizing all available information sources to predict minefield risk areas (maps) for route planning.

The current height-of-burst (HoB) scoring process is done manually by human experts. This is a time-consuming process of pausing video feeds of the tests at the burst, then estimating the burst height from camera metrics and screen fiducials. This slow and tedious process can be fully automated using well known AI/ML computer vision (CV) algorithms. The detection of the burst and subsequent height estimation can be fully automated using modern AI techniques with relatively low computational requirements, providing fast results with minimal cost in man-hours.

HoBES (Height of Burst Estimation System) is an end-to-end automated software system for burst detection and height-of-burst calculation. It supports automatic detection of the burst from a video file, utilizing a change detection approach which makes use of multiple computer vision algorithms to process and detect the burst. HoBES then processes the image of the burst frame from the video file from one or multiple angles for a height-of-burst calculation done either through XGBoost or a multi-camera triangulation approach. HoBES revolutionizes fuze testing, minimizes man-hours, and improves accuracy of height estimation using AI algorithms and other proven approaches in real world scenarios. CSLabs' open and modular DeltaX software framework, developed during an SBIR Phase II and transitioning Phase III, provided an easy-to-use configurable processing pipeline and image processing framework. Its processing pipeline has been developed to integrate with different camera systems and facilitates integration of generic AI algorithms (avoiding vendor lock-in). This framework for data processing was used in HoBES for the research and development of algorithms.

Untrusted ICs may have malicious hardware modifications modifying intended functionality. Current post silicon logical testing techniques are inadequate. Functional tests cannot discover well-disguised logical modifications, the search space is too large. Methodologies employing techniques to image metal interconnection and/or gates, can be employed in combination with functional test regimes, increasing the overall (joint) probability of detection. Post silicon design verification requires destructive de-packaging and reverse engineering, but of these, the only verification approach that is physically non-destructive and repeatable is X-ray imaging. Current X-ray imaging tools and techniques under development in programs such as IARPA's RAVEN and DMEA seek to develop brighter imaging sources, create better imaging techniques, improve optics, and scalability with respect to transistor density.

CSLabs' developed and utilized the DeltaX framework to conduct a two-layered approach to perform IC X-ray inspection: 1) macro-scale inspection, a 2D imaging approach performing inspection of a small set of X-ray images to detect changes between candidate ICs and a trusted reference; and 2) 3D computer tomography (CT) inspection. ICs were to be compared against a reference; derived from GDSII computer aided designs (CAD) or from statistically sampled candidate ICs cued from macro-scale inspection. Under the Assured and Trusted Microelectronics Solutions (ATMS) BAA (Broad Agency Announcement), FA8650-18-S-1201, CSLabs validated the efficacy of employing photon flux-limited X-ray microscopy, tomography, and macro-scale inspection to develop a novel statistical capability that supports an effective integrated circuit (IC) trust and assurance strategy.

Printed Circuit Boards (PCB) are manufactured by placing conductive electrical traces (primarily Copper) on a non-conductive substrate such as FR-4, a woven fiberglass material that also provides structural integrity. Multiple electrical layers can be "sandwiched" together with inter-layer connections made by Vias which are holes drilled through each layer and filled with conductive material to create a multi-layer PCB. After an electrical circuit is designed the layout is created with a Computer Aided Drafting (CAD) tool and manufacturing instructions for the PCB are exported in the form GERBER files. These GERBER files are provided to the PCB manufacturer for production of the PCB. In virtually all instances, the internal layers of the PCB are not visible, which can create challenges to the PCB Designer as well as End Users. The Designer must trust that the manufacturer has followed proper manufacturing practices and has kept the unit free from manufacturing defects, Foreign Object Material (FOD), or clandestine efforts. The lack of transparency also presents challenges to the End User when a PCB fails. With no way to troubleshoot the failure, the End User must replace the PCB, which can be extremely expensive particularly for aging military or aerospace platforms.

To address the mission need, CSLabs developed its XDLayer technology, a software framework designed to automate sample scanning and optimize 2D and 3D image analysis algorithms. XDLayer maximizes overall system throughput by reducing the number of viewing angles and exposure times required to analyze a sample, while incorporating real-time scanning feedback and image analysis to ensure high-quality reconstruction. By minimizing exposure times, the framework reduces the risk of radiation damage that could impair the functionality of circuits on a PCB, with the possibility of restoring functionality depending on the circuit's inherent radiation tolerance properties.

Traditional active 3D imaging systems, such as airborne and terrestrial LiDAR scanners, use a transmitter and receiver typically co-located on the same platform or connected through synchronous communications (tethered). Where, linear LiDAR imaging systems are deployed from aerial vehicles traveling at speeds of about 90 knots, an altitude of 3000 feet, collecting 8 points per meter over a swath of 3000 feet on the ground. Advanced imaging solutions such as Harris Corporations’ Geiger-mode LiDAR employ an array of sensors generating up to 204,000,000 pulses per second. Where, the Geiger-mode LiDAR employs a conical path, deployed at an altitude of 29,000 feet from an aerial vehicle traveling at 290 knots generating point densities of up to 100 points per square meter (ppsm).

Recent advances in LiDAR, detector, and airborne systems technology have opened the door to small, high-performance, and significantly lower-cost alternatives over currently deployed airborne LiDAR imaging systems. CSLabs’ initiative leveraged modeling and simulation (M&S) to evaluate the efficacy of new approaches with the potential to disrupt the existing LiDAR imaging paradigm. Where, candidate solutions employ low-cost LiDAR system components capable of functioning asynchronously while being deployed in bi/multi-static configurations to support existing and novel concepts of operations.

As semiconductor devices became smaller and more complex, the need to produce high-resolution three-dimensional digital reconstructions of integrated circuits (ICs) grew increasingly critical. Such imaging was essential for verifying ICs manufactured by untrusted foundries and for reverse-engineering obsolete or unidentified components. Achieving this required an automated framework capable of processing large volumes of raw imaging data to create detailed 3D representations with nanometer-scale precision and rapid reconstruction times. Traditional imaging technologies, including electron microscopes, focused ion beams, and laser ablation systems, faced limitations in speed, destructiveness, and repeatability. The mission need centered on developing advanced methods that could deliver high-resolution, repeatable imaging of ICs using non-destructive techniques such as X-ray computed tomography (CT). This demanded sophisticated stitching, reconstruction, and segmentation algorithms to accurately align, interpret, and map circuit features across multiple layers, enabling complete and reliable three-dimensional visualization of microelectronic structures.

To aid in producing these digital reconstructions, CSLabs developed the XRadIC framework to non-destructively analyze cutting-edge microelectronics using photon flux-limited X-ray microscope systems. The framework coupled on-the-fly sample scanning with online optimized 2D and 3D image analysis algorithms to maximize overall system throughput by dynamically minimizing the number of viewing angles and exposure times needed to analyze a given integrated circuit, while simultaneously scanning and performing image analysis of the circuit's progress. XRadIC formed the basis of a scalable, extensible platform for failure analysis and reverse engineering. Based on rigorous physics-based simulation, XRadIC's algorithm development included image stitching, limited-view computed tomography, and 3D segmentation and classification for integrated circuit wiring structures. A key innovation goal of XRadIC was to combine iterative computed tomography methods inspired by compressive sensing with segmentation methods, using as a prior the expected structures of integrated circuits such as planar wiring and binary density classification. This optimization enabled significant reductions in sample analysis times compared with conventional computed tomography algorithms on flux-limited microscope systems.

The U.S. military requires advanced RF propagation modeling and simulation capabilities to address the challenges posed by foliage-rich environments. These environments degrade RF signals, limiting access to critical information for decision-makers and warfighters. Existing models are constrained to narrow frequency ranges and fail to accurately represent signal behaviors across diverse terrains such as jungles, forests, and urban areas. The need is for a robust, scalable modeling framework that can operate across the RF spectrum and adapt to evolving military communications technologies including MIMO, UWB, and millimeter-wave systems.

CSLabs developed the Foliage Abstraction RF Multiscale Estimation Rendering (FARMER) methodology to overcome the limitations of traditional RF modeling. FARMER applies an innovative multiscale calculation approach that dynamically adjusts computational resources and modeling fidelity based on scenario complexity. It integrates acceleration algorithms, parallel processing, and adaptive selection of propagation models to deliver accurate predictions of signal behavior across a broad frequency range (up to 300 GHz). FARMER supports standalone use or modular integration into existing modeling frameworks and is applicable to military RF systems, radar, SAR, sensor networks, and more. The result is a highly flexible and powerful solution capable of simulating realistic RF behavior in dense, cluttered environments.

The client needed a reliable way to improve rental underwriting decisions by accurately assessing tenant credit risk. Traditional methods struggled with incomplete data, survivor bias, and imbalanced datasets, leading to inaccurate predictions and higher financial risk.

CSLabs played a pivotal role in supporting a successful Inc. 500 SaaS analytics company by curating and analyzing client rental and credit histories. This data was instrumental in developing advanced Deep Learning credit-risk models for rental underwriting. The Artificial Neural Network (ANN) harnessed various features, including demographic data, credit scores, and applicant information, to accurately predict the risk associated with underwriting rental agreements. To ensure the highest quality of analytics, CSLabs diligently addressed challenges like survivor bias and dataset imbalance, resulting in improved model accuracy and robustness. This enhanced the accuracy and robustness of credit-risk prediction models for rental underwriting, improved handling of survivor bias and dataset imbalance to ensure fairer, more reliable analytics, enabled the client to make faster, data-driven underwriting decisions with greater confidence, and strengthened the overall underwriting process, reducing financial risk and improving portfolio performance.

The Army requires an advanced and user-interactive capability to efficiently process Wide-Area Motion Imagery (WAMI) datasets for real-time vehicle tracking and classification. As the scale and complexity of WAMI data increases, there is a critical need for a system that enables both automated processing and human-in-the-loop validation to improve accuracy, reduce operator workload, and accelerate decision-making in dynamic operational environments.

CSLabs supported Korb Satellite Systems on an ARMY contract to develop a tracking and classification application for processing Wide Area Motion Imagery (WAMI) datasets. CSLabs developed an interactive GUI that enabled operators to rapidly import and process large-scale WAMI datasets through an intuitive interface, streamlining analysis workflows for time-sensitive military operations. Users could configure and save processing options to a configuration file for reuse, annotate a region-of-interest (ROI) for backend tracking, run detection and tracking processes, and interact with annotated images to modify and correct results, such as missing vehicle tracking links. Integrated backend modules detected and tracked multiple moving vehicles across wide-area scenes, reducing the need for manual frame-by-frame review and supporting operators with efficient, accurate WAMI dataset processing.

To address the overwhelming scale and complexity of modern patent research, there is a critical need for an AI-enabled system that can efficiently process, interpret, and retrieve relevant patent information from rapidly growing datasets that include both textual descriptions and technical figures. Current methods are insufficient to handle the exponential growth of data and evolving terminology in the intellectual property space.

CSLabs, in partnership with Matt Kasap Inc. (MK Inc.), developed the SMart AI Research Tool (SMART), an AI-enabled patent research capability created to address the exponential growth of information and terminology faced by researchers. The system was designed to transform patent search and retrieval by integrating multi-modal Large Language Models (LLMs), reinforcement learning, and a hierarchical multi-model AI architecture to improve both precision and efficiency. Unlike traditional tools that relied solely on text, SMART processed both textual and visual data, including figures and diagrams, contained in patent applications. This multimodal capability was extended into a scalable, adaptable framework that allowed the system to learn and refine strategies across multiple models and disciplines. SMART reduces the time required to locate relevant prior art and technical disclosures, improves precision and recall by integrating text and imagery into the search process and demonstrates the ability to adapt across industries and patent classes with tailored search strategies for diverse technical fields.