AMOS Recap: Revolutionizing Space Situational Awareness Through Applied AI & ML
Last month, the Slingshot team traveled to Hawaii for the 25th Advanced Maui Optical and Space Surveillance Technologies (AMOS) Conference, the premier industry event focused on advancing space situational awareness (SSA) for the global space community. It was a week rich with learning, collaboration, and engagement with leading customers and industry partners at the forefront of SSA innovation.
During the conference, our team had the privilege of presenting three technical papers highlighting Slingshot's cutting-edge AI/ML SSA technologies. These groundbreaking solutions are designed to tackle the escalating challenges posed by the increasingly dynamic space environment. Below, we preview each paper, including the key innovations and how they are set to transform the future of SSA.
Links to the full papers below!
1. Seeing the Unseen: ML-Based Photometric Fingerprinting for LEO Satellite Monitoring
The Challenge: Traditional astrometric observation methods struggle to differentiate space objects and detect potential hazards in the crowded low Earth orbit (LEO) environment.
The Innovation: This paper introduces a two-step machine learning approach that exploits the high volume of photometric (brightness) data generated by our Horus optical fence systems to create unique "fingerprints" of satellites.
Key Advancements:
- Novel photometric fingerprinting algorithm enabling object characterization, identification, and classification
- Methods for embedding and comparing fingerprints that will soon help to detect satellite attitude changes
- Demonstrated performance using real LEO observations
Practical Impact: Imagine being able to identify a specific satellite among thousands, detect subtle changes in its behavior, or even infer its purpose based on its "light signature." This technology could revolutionize threat detection, space traffic management, and satellite health monitoring.
🏆 Download the paper that our first-place winning poster was based off of: ML-Based Photometric Fingerprinting for LEO Satellite Monitoring
2. Predicting the Unpredictable: Early Identification of High-Covariance Conjunctions
The Challenge: Satellite operators face a constant dilemma: when to commit resources to potential collision avoidance maneuvers in an uncertain environment.
The Innovation: Slingshot Aerospace has developed a machine learning model that predicts which conjunction events are likely to require additional data acquisition, days before a decision deadline.
Key Advancements:
- Predicts high-covariance conjunctions up to five days before closest approach time
- Incorporates temporal evolution of Conjunction Data Message (CDM) sequences
- Integrates contextual information about objects involved in conjunctions
- Outperforms baseline covariance-only approaches
Practical Impact: This predictive model could reduce false alarms and unnecessary maneuvers, saving fuel and extending satellite lifespans. It gives operators crucial lead time to request additional observations and make informed decisions.
Download: Contextual Predictive Model for Early Identification of High-Covariance Conjunctions
3. Spotting the Odd One Out: Action-Free IRL for Satellite Similarity and Anomaly Detection
The Challenge: As low Earth orbit mega-constellations become the norm, detecting anomalous behavior and characteristics amongst thousands of similar satellites is like finding a needle in a haystack.
The Innovation: This paper presents a novel action-free Inverse Reinforcement Learning (IRL) approach to characterize inter-satellite similarity and identify anomalies.
Key Advancements:
- Action-free IRL method for analyzing satellite behavior without known action data
- Scalable ML capabilities for processing large collections of resident space objects (RSOs)
- 15-20x computational speedup compared to dynamic time-warping methods
- Demonstrated ability to distinguish anomalies in a simulated LEO constellations
Practical Impact: This technology could be a game-changer for constellation management and space domain awareness. It enables rapid identification of satellites operating outside normal parameters, whether due to malfunction, cyber attack, or intentional maneuvering. A real-life example can be found HERE.
Download: Action-Free Inverse Reinforcement Learning for Satellite Similarity and Anomaly Detection