Quantum-Inspired Software Cuts AI Model Size by 99% to Speed Up Identification of Unknown Orbital Objects

Quantum-Inspired Software Cuts AI Model Size by 99% to Speed Up Identification of Unknown Orbital Objects

New York-based technology firm BosonQ Psi Federal (BQP) has won its first U.S. federal research contract, tasked with developing software to help military operators more rapidly identify unknown objects in Earth orbit. The contract was awarded through the SpaceWERX Open Topic Small Business Innovation Research (SBIR) program, with a focus on leveraging physics-constrained machine learning to enhance Space Domain Awareness (SDA).

BQP will validate a novel software application that combines physical modeling with quantum-inspired computing, with the goal of classifying unidentified orbital objects faster and at a fraction of the computational cost of conventional AI models. The technology is designed specifically for satellites and other edge platforms operating under strict power and processing constraints.

A More Crowded Orbital Environment

As the number of satellites and pieces of space debris continues to grow, tracking orbital activity has become increasingly challenging. The U.S. Space Surveillance Network collects between 18,000 and 25,000 observations per day, many of which cannot be immediately correlated with known satellites or debris.

These unmatched detections are known as Uncorrelated Tracks (UCTs) and may include newly launched spacecraft, collision fragments, or objects requiring further investigation. Delays in identification slow operational decision-making and degrade overall situational awareness of the space environment.

BQP's software aims to accelerate this identification process by combining physical constraints with quantum-assisted machine learning. The company says the approach delivers accurate AI inference without relying on cloud infrastructure, graphics processing units (GPUs), or future quantum computers—running directly on space-qualified processors and other resource-limited hardware.

Smaller AI Models, Faster Decisions

According to BQP, its Physics-Constrained Quantum-Assisted Machine Learning (PC-QAML) architecture compresses models by 99%—from approximately 14 million parameters down to around 2,000—without sacrificing accuracy. The company reports that classification accuracy remains above 99% despite the dramatic reduction in model size.

The compact architecture also delivers up to a 10× reduction in inference latency and approximately 90% savings in power consumption. BQP adds that engineers can retrain the models significantly faster than with conventional machine learning systems.

These performance gains have been validated through deployment on NVIDIA Jetson Nano edge computing devices at the Space Domain Awareness TAP Lab (formerly the SDA TAP Lab), demonstrating that advanced AI can run on compact hardware suitable for autonomous space missions.

BQP founder and Chief Technology Officer Rut Lineswala said: "Our goal is to make advanced AI useful where it matters most—on satellites and forward-deployed systems operating with limited compute and intermittent connectivity."

He noted that the federal contract validates the company's technology and provides an opportunity to demonstrate how quantum-inspired computing can address operational challenges faced by national security missions.

Applications Beyond Military Operations

Military operators could use the software to distinguish routine orbital activity from potentially suspicious behavior, including satellite maneuvers, separation events, and close-proximity operations. Processing data directly onboard spacecraft reduces dependence on centralized computing systems and improves response times.

The program builds on BQP's earlier collaboration with the Space Domain Awareness TAP Lab. During the 2025 SDA Mini-Accelerator, BQP's technology demonstrated orbital separation detection capabilities and was identified as a candidate for future UCT classification and threat simulation applications in support of U.S. space operations.

Beyond defense, the same software has potential applications in commercial aerospace, autonomous vehicles, and industrial monitoring—anywhere AI needs to run reliably on small, low-power computing platforms.

原文來源： 查看原文