NUWA / SISP — self-healing satellite constellations
2nd of 193 teams at the international IEEE AESS Sustainability Hackathon, with an anomaly detector validated on ESA telemetry at 0.99 accuracy.
Because the sky shouldn't wait for someone on the ground to fix it.
- 273/273
- Assertions passed
- 94.3 %
- RMSE improvement
- 0.99
- Detector accuracy
- 9 dB
- Link margin at ~1000 km
- 100 %
- Payload reconstruction
Context
Low Earth orbit carries more than 1.2 million debris objects larger than 1 cm, and 41.3% of small satellites launched between 2000 and 2016 suffered total or partial mission failure. A single ground station gets two to four usable contacts a day, minutes each.
So when a sensor starts degrading, the spacecraft waits. Often it is decommissioned early — and becomes debris itself, making the problem worse. The premise of NUWA is that the fix belongs in software already in orbit, not in more backup hardware launched to replace it.
Approach
- Correction — a DEGR-weighted median-Kalman filter repairs degraded sensor readings in place, so a partially failed instrument still yields usable data.
- Relay — vital data is routed through healthy neighbours when a direct link to ground is unavailable.
- Borrow — a failed sensor is bypassed entirely by streaming a neighbouring satellite's healthy readings in its place.
- All three are governed by DEGR, a trust score from 0 to 15 deciding how much weight a satellite carries in consensus — from full participation, through reduced and near-zero weight, to exclusion.
In plain terms
A way for a group of small satellites to notice when one of them is going blind and cover for it — correcting its readings, relaying its data, or lending it their own sensors — instead of letting it become space junk.
The data path
Telemetry enters at the left, the detector scores it, and that score becomes the weight every downstream service reads. Current runs the live path: sensors through detection to the state machine, and back down into the three services it governs.
Five satellites at DEGR 1 through 15, from healthy to critical failure. Below, the SVD anomaly pipeline and the protocol stack; to the right, the trust bands that decide whether a node's data still counts in consensus.
Constellation health, link matrix and a live protocol log, with fault, low-energy and ground-link scenarios injectable at runtime. Validated against the European Space Agency's OPSSAT-AD telemetry dataset.