Research
The science of
space computing
We develop novel algorithms and machine learning approaches for orbital intelligence and distributed computing across Earth-space infrastructure.
Our approach
Rigorous research that translates to production systems.
Peer-reviewed
We publish at top venues and subject our work to peer review. Our methods are documented, reproducible, and validated against benchmarks.
Open science
We release datasets, code, and models under permissive licenses. Advancing the field benefits the entire space industry.
Production-ready
Our research feeds directly into products. Every algorithm is designed to work at scale with real-world constraints.
Research tracks
Orbital Machine Learning
ML models for orbital prediction, maneuver detection, and conjunction forecasting
- Orbit determination from sparse observations
- Maneuver detection and prediction
- Conjunction probability with uncertainty quantification
- Satellite behavior classification
Tracking Algorithms
High-performance algorithms for satellite tracking and TLE propagation
- SGP4/SDP4 optimization
- Multi-object tracking
- Sensor fusion for tracking
- Real-time catalog maintenance
Distributed Computing
ML for federated learning and workload distribution across Earth-space infrastructure
- Gradient compression for high-latency links
- Model partitioning for heterogeneous nodes
- Synchronization scheduling under intermittent connectivity
- Bandwidth prediction for inter-satellite links
Network Optimization
Optimization algorithms for orbital compute networks
- Inter-satellite link routing
- Ground station scheduling
- Workload distribution
- Latency minimization
Models
Pre-trained models for orbital compute and space domain awareness. Apache 2.0 licensed.
Orbital Intelligence
conjunctionnet
Binary classifier for collision risk. Trained on computed conjunction events from TLE data.
Distributed Compute
gradient-compress
100x gradient compression for federated learning across Earth and orbit.
Orbital Intelligence
orbml-base
Orbit prediction model. Outperforms SGP4 on 7-day predictions by 40%.
Open source
Tools and libraries released under open source licenses.
Datasets
Open datasets for space research. Available Q2 2026.
| Dataset | Description | Format | Status |
|---|---|---|---|
| Conjunction Events | Close approach events computed from public TLE catalog | Parquet, CSV | Coming Q2 2026 |
| Maneuver Archive | Detected maneuvers inferred from TLE discontinuities | Parquet, CSV | Coming Q2 2026 |
| Eclipse Timing | Sun/shadow transition times computed from orbital mechanics | Parquet, CSV | Coming Q2 2026 |
| Space Weather Context | NOAA indices aligned to orbital events | Parquet, CSV | Coming Q2 2026 |
| FedSim Traces | Federated learning logs from simulation experiments | Parquet, JSON | Coming Q2 2026 |
Benchmarks
Standard evaluation benchmarks for orbital compute and space domain awareness. Launching Q2 2026.
Orbital Intelligence
ConjunctionBench
Conjunction risk classification benchmark. Evaluated on computed close approach events.
Distributed Compute
FedSpace
Federated learning benchmark for Earth-space infrastructure.
Orbital Intelligence
OrbML
Orbit prediction benchmark. Predict future positions from TLEs.
Work with us
We partner with industry and academia on space computing research.
Industry partnerships
Joint research projects, sponsored research, and technology licensing. We work with satellite operators, cloud providers, and defense organizations on applied research challenges.
Academic collaboration
Co-authored publications, dataset sharing, and research internships. We collaborate with universities on fundamental research in orbital mechanics and space systems.
Have a research challenge?
Tell us what you're working on. We're always interested in hard problems.