FOR IMMEDIATE RELEASE | September 10, 2025 – Carnegie, PA
NASA SBIR Phase I project will support development of MVPT, a model-agnostic approach for real-time validation of machine learning systems operating on novel data in autonomous space missions.
VISIMO Wins NASA SBIR Phase I Award to Develop MVPT for Real-Time Validation of Autonomous AI Systems
CARNEGIE, PA. – VISIMO, a Carnegie-based artificial intelligence and software engineering company, has been awarded a NASA Small Business Innovation Research (SBIR) Phase I project to develop Model Validation via Precomputed Transformations (MVPT), a new approach designed to help autonomous systems rapidly validate machine learning models when they encounter new, previously unobserved data in mission-critical environments.
Autonomous systems play a critical role across NASA missions because they can operate in extreme environments, function for long durations with minimal human intervention, and support science and operations far beyond the reach of direct human control. As NASA increasingly relies on machine learning to power perception, decision support, anomaly detection, and other advanced capabilities, model validation becomes a central mission assurance problem.
That challenge is especially difficult in space operations because many autonomous systems are deployed into environments where the most important data does not exist before launch. Traditional validation methods are typically grounded in previously collected training data or time-intensive revalidation workflows, making them difficult to apply when a model must adapt safely and quickly to novel conditions during a mission.
MVPT is designed to address that problem by shifting much of the validation burden to the pre-mission phase. Rather than treating validation as a static checkpoint, MVPT precomputes how a model’s validation status changes across a broad transformation space derived from the training data. When new mission data is encountered, the system can map the incoming data to the closest transformation state and rapidly determine the model’s expected validation status in real time.
“Reliable autonomy depends on more than building capable models. It depends on understanding whether those models can still be trusted when conditions change,” said James Julius, Founder and CEO of VISIMO. “MVPT reflects the kind of mission-focused AI work VISIMO is built for: creating practical, defensible tools that help operators manage uncertainty, reduce risk, and protect mission success.”
Advancing Mission-Safe Autonomy
MVPT is intended to be model-agnostic and data-modality agnostic, making it broadly applicable across a wide range of NASA-relevant machine learning systems. The approach is designed to support images, radar, time-series data, and other mission data types, and it can be paired with different validation methodologies rather than being limited to a single modeling framework.
This flexibility gives MVPT the potential to support robotic exploration platforms, remote sensing systems, future autonomous spacecraft operations, mission science workflows, and commercial applications where AI systems must perform safely under unfamiliar conditions.
Reducing Risk, Cost, and Operational Latency
By enabling faster validation decisions on novel data, MVPT is intended to improve both mission resilience and mission economics. The approach can help identify model weaknesses before launch, reduce the operational burden of revalidating systems in flight, and support more responsive autonomous behavior when delays or limited communications make traditional human-in-the-loop workflows impractical.
The ability to shift computationally intensive analysis into the pre-mission phase also creates opportunities to leverage powerful ground-based computing resources while minimizing the onboard burden placed on mission hardware. In practice, that can support safer deployment of advanced AI capabilities without requiring spacecraft or remote systems to carry the full cost of exhaustive real-time validation.
Phase I Technical Objectives
During Phase I, VISIMO will evaluate MVPT’s technical feasibility through three core objectives: assessing the compatibility of the approach across relevant machine learning models, validation methods, data modalities, and data transformations; documenting the algorithmic approach in detail and demonstrating it on a simplified working example; and establishing the Phase II simulation plan needed to validate the method in a NASA-relevant scenario.
The Phase I effort is intended to lay the foundation for a Phase II simulation that more fully tests MVPT against realistic mission conditions, including changing inputs, noisy measurements, model adaptation, and the operational consequences of correct or incorrect validation outcomes.
“MVPT is about giving autonomous systems a faster and more rigorous way to reason about trust when they encounter the unknown,” said Dr. Alexander Moskowitz, Principal Investigator and Principal Research Scientist at VISIMO. “If we can help mission teams understand how model validation changes under novel conditions before those conditions occur, we can materially improve the safety and effectiveness of AI-enabled exploration.”
About VISIMO
VISIMO is a Carnegie, Pennsylvania-based AI and software engineering company founded in 2015. VISIMO builds applied machine learning, data science, and mission software solutions for federal agencies and commercial partners, with a focus on rapid prototyping, defensible AI, and decision-support systems. VISIMO specializes in transforming complex data and technical uncertainty into practical tools that support better decisions in high-consequence environments.
Media Contact:
VISIMO Media Inquiries
VISIMO
412-699-6900
media@visimo.ai
Government Disclaimer:
This material is based upon work supported by the National Aeronautics and Space Administration. Any opinions, findings, conclusions, or recommendations expressed in this material are those of VISIMO and do not necessarily reflect the views of NASA.