2 Sep 2026

Explainability-Driven LangChain-Integrated Large Language Models and Knowledge Graphs for Multiagent Reinforcement Learning in Complex Air Combat Simulation

Abstract: This work advances multiagent reinforcement learning (MARL) for complex air combat by integrating methods that enhance decision-making and explainability. Using a realistic six-degrees-of-freedom aerial simulation built on the OpenAI Gymnasium framework, we investigate competitive agent interactions in dynamic scenarios. We apply explainability techniques to clarify agent behavior and interaction patterns. The MARL framework is further augmented with knowledge graphs, large language models, and modular orchestration using the LangChain framework. This combines data-driven learning with knowledge-driven reasoning to strengthen situational awareness, coordination, and interpretability. Experimental results indicate that this integration sustains competitive performance while enhancing transparency and human interpretability.

1 Sep 2026

Evaluation of Air Combat Simulation for Multiagent Reinforcement Learning Approaches

Abstract: Reinforcement learning (RL) is increasingly used to develop intelligent control policies for complex and dynamic systems such as air combat simulations. This report reviews several open-source frameworks, including JSBSim, FlightGear, CyberBattleSim, CloseAirCombat, and DogFight2, with a focus on their suitability for RL-based aircraft simulation and multiagent research. The evaluation considers each platform’s flexibility, ease of integration, and support for various RL methods. Based on this analysis, the report proposes a modular simulation framework that combines accurate flight dynamics with standardized RL interfacing. This work aims to accelerate research on autonomous air combat agents and provide a foundation for future advancements in multiagent RL.

19 Aug 2026

A Hybrid Statistical–Dynamical Framework for Compound Coastal Flooding Analysis

Abstract: Compound coastal flooding due to astronomic, atmospheric, oceanographic, and hydrologic drivers poses severe threats to coastal communities. While physics-driven approaches are able to dynamically simulate temporally and spatially varying compound flooding generated by multiple drivers with correlations between some of them, computational burdens limit their capability to explore the full range of conditions that contribute to compound coastal hazards. Data-driven statistical approaches address some of these computational challenges; however, they are also unable to explore all possible forcing combinations due to short observational records, and projections are typically limited to a few locations. This study proposes a hybrid statistical–dynamical framework for compound coastal flooding analysis that integrates a stochastic generator of compound flooding drivers, a hydrodynamic model, and machine learning-based surrogate models. The framework was demonstrated in San Francisco Bay over the past 500 years with accuracy similar to the physics-driven approach but with much higher computational efficiency. The stochastic generator of compound flooding drivers is developed by coupling a sea surface temperature reconstruction model with a climate emulator, weather generator, and model of the hydrological and reservoir system. Using reconstructed SSTs as input, the generator of compound flooding drivers is employed to simulate time series of the forcing factors contributing to compound flooding in SF Bay. A process-based hydrodynamic model is built to predict total water levels varying in time and space throughout SF Bay based on stochastically generated drivers. The machine learning-based surrogate models are then developed from a relatively small library of hydrodynamic model simulations to efficiently predict water levels for compound flooding analysis under the full range of stochastic drivers. This study contributes a hybrid statistical–dynamical framework to better understand the spatial distribution and temporal evolution of compound coastal-fluvial flooding, along with the relative contributions of drivers in complex nearshore, estuarine, and river environments for centennial timescales under past, present, and future climates.

10 Jul 2026

Microsoft Azure Artificial Intelligence / Machine Learning Hackathon for Development of Retrieval-Augmented Generation Large Language Model

Abstract: The US Army Corps of Engineers (USACE) Civil Works (CW) research and development (R&D) mission is to address challenging environmental sustainability problems through innovative science and engineering, which helps to ensure a safer, more prosperous, and more resilient nation. To achieve this, the US Army Engineer Research and Development Center (ERDC) plans, executes, leads, and directs many R&D programs in coordination with USACE Headquarters, Districts, and Divisions through its multiple strategic focus areas, which include infrastructure, water modeling, crisis preparedness, ecosystem, sediment management, data, artificial intelligence, and robotics. In this process, much information is generated, including internal progress reviews, financial reports, scopes of work, work package planning, and success stories.

29 Apr 2026

Deep Learning Approaches for Buried Object Detection in Infrared Imagery

Abstract: Artificial intelligence and machine learning techniques are increasingly utilized to detect buried objects in thermal infrared imagery. This task relies heavily on the quality and diversity of the training dataset, requiring datasets that capture variability present in real-world environments. Synthetic imagery offers a means to expose algorithms to a greater range of conditions than is often available in real-world data alone. This study evaluates the performance of three open-source object detection models—Faster Region-Based Convolutional Neural Network (R-CNN), You Only Look Once (YOLOv8), and Single Shot Multibox Detector—trained using real-world, synthetic, and hybrid datasets. Real-world imagery was collected from a single field site, while synthetic data were generated using the Virtual Environmental Simulation for Physics-Based Analysis software suite. Model performance was evaluated using Intersection over Union and confidence scores. Models trained exclusively on synthetic datasets with limited scene diversity, when tested on real-world imagery from the same location, produce high false-positive and false-negative rates. Detection performance im-proved significantly for Faster R-CNN and YOLOv8 when trained using a hybrid dataset combining real-world and synthetic data. Analysis of red-green-blue histograms revealed differences in pixel intensity distributions between real and synthetic imagery, indicating areas for improving synthetic data generation.

1 Oct 2025

Predicting Seagrass Habitat Suitability with Remote Sensing and Machine Learning: A Case Study in the Mississippi-Alabama Barrier Islands

Abstract: Seagrass occupies sandy platforms landward of the Mississippi-Alabama barrier islands, where the benthos experiences consistent sediment transport. This work characterized benthos surrounding Cat Island, Mississippi, to assess the influence of elevation and geomorphological features (e.g., slopes, flats, peaks, and valleys) on seagrass presence. Two machine learning algorithms predicted seagrass presence/absence (from airborne hyperspectral imagery) based on elevation and geomorphology (derived from airborne lidar bathymetry) for 2016, 2018, and 2019. Results indicated elevation primarily influenced seagrass presence, with minimal impact from geomorphology. Elevation change was not predictive, suggesting seagrass tolerated observed deposition/erosion rates. This research showcases remote sensing and machine learning efficacy in predicting seagrass habitat suitability (greater than 70% accuracy) and conveys implications for conservation.

9 Sep 2025

Using Transfer Learning to Enhance Void Detection and Shear Wave Velocity Model Inversion from Near-Surface Seismic Shot Gathers

Abstract: A Convolutional Neural Network (CNN) has been designed to delineate the shear-wave velocity (Vs) models and detect subsurface void locations. Addressing the processing and interpretation challenges posed on real seismic data, our strategy emphasizes that leveraging the ground truth, which is the void location in this study, enables the CNN to catch the identical features in real waveforms. Initially, a synthetic dataset is employed, imparting foundational knowledge to the CNN regarding the Vs model and void locations. Drawing inspiration from transfer learning, this pre-trained CNN serves as an initial model and is refined using a real dataset focused on void locations. After refining, the CNN shows enhanced reliability to detect the void and extract the Vs model, as evidenced by the improved alignment between forward modeling and real waveforms. Our findings underscore how leveraging the ground truth can actualize the potential of CNN on velocity model extraction.

11 Jun 2025

Prediction of Waterborne Freight Activity with Automatic Identification System Using Machine Learning

Abstract: This paper addresses latency issues related to publicly available port-level commodity tonnage reports. Predicting commodity tonnage at the port-level, near real time vessel tracking data is used with historical WCS with a machine learning model. Commodity throughput is derived from WCS data which is released publicly approximately two years after collection. This latency presents a challenge for short-term planning and other operational uses. This study leverages near real time vessel tracking data from the AIS data set. LSTM, TCN, and TFT machine learning models are developed using the features extracted from AIS and the historical WCS data. The output of the model is the prediction of the quarterly volume of commodities at port terminals for four quarters in the future. Uncategorized and Categorized models were developed. The uncategorized outperformed the categorized based on the Mean Absolute Percentage Error. The uncategorized LSTM model has the highest accuracy. Results show the model has higher accuracy for port terminals that handle a specific type of vessel, compared to the port terminals handling more than one vessel type. The application of the model enables port authorities and stakeholders to make short-term capacity expansion and infrastructure investment decisions based on commodity volume.

15 May 2025

Discriminating Buried Munitions Based on Physical Models for Their Thermal Response

Abstract: Munitions and other objects buried near the Earth’s surface can often be recognized in infrared imagery because their thermal and radiative properties differ from the surrounding undisturbed soil. However, the evolution of the thermal signature over time is subject to many complex interacting processes, including incident solar radiation, heat conduction in the ground, longwave radiation from the surface, and sensible and latent heat exchanges with the atmosphere. This complexity makes development of robust classification algorithms particularly challenging. Machine-learning algorithms, although increasingly popular, often require large training datasets including all environments to which they will be applied. Algorithms incorporating an understanding of the physical processes underlying the thermal signature potentially provide improved performance and mitigate the need for large training datasets. To that end, this report formulates a simplified model for the energy exchange near the ground and describes how it can be incorporated into maximum-likelihood ratio and Bayesian classifiers capable of distinguishing buried objects from their surroundings. In particular, a version of the Bayesian classifier is formulated that leverages the differing amplitude and phase response of a buried object over a 24-hour period. These algorithms will be tested on experimental data in a future study.

25 Apr 2025

Analysis Tools and Techniques for Evaluating Quality in Synthetic Data Generated by the Virtual Autonomous Navigation Environment

Abstract: The capability to produce high-quality labeled synthetic image data is an important tool for building and maintaining machine learning datasets. However, ensuring computer-generated data is of high quality is very challenging. This report describes an effort to evaluate and improve synthetic image data generated by the Virtual Autonomous Navigation Environment’s Environment and Sensor Engine (VANE::ESE), as well as documenting a set of tools developed to process, analyze, and train models from, image datasets generated by VANE::ESE. Additionally, the results of several experiments are presented, including an investigation into using explainable AI techniques, and direct comparisons of various models trained on multiple synthetic datasets.