Research

NETS Lab organizes its work around a small number of connected research programs. Rather than treating theory, modeling, and deployment as separate activities, the lab uses each to inform the others: we build methods, stress them in realistic environments, and study what makes them reliable and interpretable.

Each section below highlights a research theme, the kinds of projects it supports, and a reverse-chronological paper list that gives a sense of the work connected to that area.

Space Perception and Autonomy

A major application area for the lab is space-based perception: inspection, tracking, navigation, and 3D understanding of spacecraft in difficult orbital environments. This includes systems that must operate onboard under strict compute, sensing, and reliability constraints, as well as methods that support rendezvous, servicing, and characterization of non-cooperative targets.

3D reconstruction is treated here as part of perception rather than as a separate theme. The same vision stack that supports detection, localization, and scene understanding also supports geometry, structure recovery, and downstream autonomy.

Research streams

Perception models. We develop core vision models for object detection, component recognition, tracking, and scene understanding in orbital imagery. This includes work on model robustness, dataset design, and architectures tailored to space-domain targets.

Inspection, pose, and 3D reconstruction. A second thread focuses on understanding spacecraft geometry and state through perception, including 3D reconstruction, pose-aware vision, and inspection workflows for non-cooperative resident space objects.

Onboard and low-SWaP deployment. We study what it takes to move these methods into flight-relevant settings, including efficient models, FPGA and embedded implementations, and comparisons across commercial off-the-shelf small computing platforms.

Autonomous experimentation. The lab also pushes toward real-world autonomy through synchronized sensing setups, experimental test stands, satellite mockups, and lab flight-style testing that connects perception to decision-making in realistic closed-loop conditions.

Current directions

• Spacecraft inspection, tracking, and component recognition
• 3D reconstruction, inspection, and pose-aware perception pipelines
• Onboard and low-SWaP vision systems for orbital missions
• Relative navigation and autonomy around non-cooperative resident space objects
• Embedded deployment and COTS small-compute comparisons
• Real-world experimental systems that bridge simulation and physical testing

2026. A. Issitt, T. Mahendrakar, and R. T. White. “Reliable Onboard 3D Reconstruction of Unknown Spacecraft via Data Acquisition Guided by Orbital Geometry and Lighting.” AIAA SCITECH 2026 Forum.

2026. N. Welsh, L. J. Shikhman, M. N. Attzs, S. K. Putane, V. M. Nguyen, and R. T. White. “Post-Launch Capability Expansion of Vision-Language Models via Prompting for On-Orbit Spacecraft Inspection.” CVPR 2026 Workshop AI4Space.

2025. A. Issitt, T. Mahendrakar, A. Alvarez, R. T. White, and A. Sizemore. “On Optimal Observation Orbits for Learning Gaussian Splatting-Based 3D Models of Unknown Resident Space Objects.” AIAA SCITECH 2025 Forum.

Explainable AI and Neural Representations

A second core theme asks what models are actually learning and how that internal structure can be controlled. The lab studies explainable AI, latent structure, entropy flow through deep networks, regularization-driven representation shaping, ensemble diversity, and methods for extracting patterns that generalize across tasks and conditions.

This area is where the lab’s mathematics background shows up most directly: not only in analyzing learned systems, but in designing objectives and constraints that guide them toward more useful internal behavior.

Current directions

• Entropy-aware training and information flow in deep networks
• Interpretable and controllable latent structure
• Diverse representations for ensembles and transfer
• Methods for identifying robust, generalizable patterns in neural representations

2026. A. Issitt, A. Merino, L. Deen, R. T. White, and M. J. Meni. “Uncovering Neural Learning Dynamics Through Latent Mutual Information.” Entropy 28(1), 118.

2025. M. J. Meni, T. Mahendrakar, O. D. M. Raney, R. T. White, M. L. Mayo, and K. R. Pilkiewicz. “Probabilistic Explanations for Entropic Knowledge Extraction for Automated Satellite Component Detection.” Journal of Aerospace Information Systems 22(4): 296-309.

2024. M. J. Meni, R. T. White, M. L. Mayo, and K. R. Pilkiewicz. “Entropy-Based Guidance of Deep Neural Networks for Accelerated Convergence and Improved Performance.” Information Sciences 681: 121239.

Digital Twins and Synthetic Sensing

The lab develops digital-twin environments to accelerate vision and autonomy research before moving into physical experiments. These environments support orbital motion, synchronized observers, controllable cameras, synthetic sensor feeds, component annotation, and mixed real/sim pipelines that help the lab iterate quickly while preserving experimental realism.

This work connects modeling and deployment directly: the same systems that support controlled synthetic experiments are also used to inform test stands, satellite mockups, and real-sensor experiments.

Current directions

• LEO-to-GEO digital twins for orbital motion and observation
• Controllable synthetic sensor feeds and synchronized observers
• Auto-annotation pipelines for detection and segmentation
• Mixed real/sim sensing and physical testbed development

2026. A. Issitt, T. Mahendrakar, and R. T. White. “Reliable Onboard 3D Reconstruction of Unknown Spacecraft via Data Acquisition Guided by Orbital Geometry and Lighting.” AIAA SCITECH 2026 Forum.

2026. A. Issitt, E. Happy, E. Clark, M. J. Meni, and R. T. White. “Assessing the Predictive Value of Physics-Grounded Synthetic Data for Computer Vision in Space Environments.” Submitted to CVPR 2026 Workshop SynData4CV.

2025. A. Issitt, T. Mahendrakar, A. Alvarez, R. T. White, and A. Sizemore. “On Optimal Observation Orbits for Learning Gaussian Splatting-Based 3D Models of Unknown Resident Space Objects.” AIAA SCITECH 2025 Forum.

Scientific and Engineering Machine Learning

NETS also works on collaborative machine learning projects in science and engineering beyond the space domain. These efforts are typically partner-driven and involve adapting methods to settings where data are messy, goals are domain-specific, and interpretation matters as much as predictive performance.

Examples include work in aerospace engineering, aviation requirements and safety analysis, computational biology, coastal and environmental systems, and other data-rich scientific settings.

Current directions

• Domain-adapted machine learning for science and engineering workflows
• Natural language models for aviation and aerospace requirements analysis
• Applied learning systems for biology, environmental science, and related areas
• Collaborative projects that connect mathematical insight to deployment constraints

2025. M. P. Cote, M. E. Splitt, S. M. Lazarus, R. T. White, and C. G. Baker. “Ground-Based Cloud Type Classification for Aviation Weather Hazard Detection Using Deep Learning.” IEEE Access 13: 203027-203040.

2025. R. E. White, M. Mattei, B. Diaz, S. Kaden, M. Brenner, E. Smith, M. A. H. Khan, et al. “Reconstruction of 3D Vascular Flow Patterns from Sparse Angiographic Data Using a 3D Convolutional Neural Network.” Medical Imaging 2025: Clinical and Biomedical Imaging 13410: 284-294.

2025. M. Mattei, A. Issitt, M. A. Khan, B. Diaz, R. White, and V. Chivukula. “Predicting Cardiac Flow Patterns Using a Novel Hemodynamic Neural Network for LVAD Therapy Planning and Evaluation.” The Journal of Heart and Lung Transplantation 44(4): S133-S134.

Applied Probability and Stochastic Systems

The broader research program also includes earlier and ongoing work in applied probability, stochastic analysis, random walks, reliability, queueing, and related mathematical systems. This work remains part of the intellectual foundation of the lab, especially in how it approaches uncertainty, dynamics, and mathematically structured modeling.

2024. J. H. Dshalalow, H. Aljahani, and R. T. White. “Dependent Competing Failure Processes in Reliability Systems.” Entropy 26(6): 444.

2022. J. H. Dshalalow and R. T. White. “Fluctuation Analysis of a Soft-Extreme Shock Reliability Model.” Mathematics 10(18): 3312.

2022. R. T. White. “On the Exiting Patterns of Multivariate Renewal-Reward Processes with an Application to Stochastic Networks.” Symmetry 14(6): 1167.

Representative Projects

Across these themes, the lab runs a mix of long-horizon research threads, project-based collaborations, and experimental system-building efforts. The projects below reflect the kinds of integrated problems NETS is designed to tackle.

All Publications

This page serves as both the research overview and the working publications hub for the lab. For the most complete and continuously updated list, see Google Scholar.

• Google Scholar
• ORCID