OPS-SAT: models trained and run on board

Trained a fault-detection model in flight from the satellite's own sensor stream, and classified and clustered its own images.

L3 Launched: confirmed by a party with no stake, or by public data. Rests on c6.
R2 Ran: own technical account. Rests on c3, c4, c5.

Operator	European Space Agency
Developer	The OPS-SAT flight control team (Labrèche and colleagues)
Host	OPS-SAT
Where	Earth orbit
Class	learning
Hardware	not stated in what was read
Model or software	Online machine learning algorithms for training; TensorFlow Lite for inference; k-means for clustering
Launched	18 Dec 2019
Ran	by April 2021
Ended	22 May 2024
Catalogue	OPS-SAT, number 44878; matched by name, the reviewer's judgement

What was said, and by whom

“To our knowledge, this is the first time an artificial intelligence model is trained onboard a flying mission.”

Georges Labrèche, experimenter: the experimenter. First Machine Learning Models Trained in Space, Georges Labrèche (personal site of the experiment's lead). Published 19 Apr 2021 (page metadata). Read 4 Oct 2026, in full.
“One of our models already gave us predictions with a whopping 89% balanced accuracy after only a few runs.”

Georges Labrèche, experimenter: the experimenter. First Machine Learning Models Trained in Space, Georges Labrèche (personal site of the experiment's lead). Published 19 Apr 2021 (page metadata). Read 4 Oct 2026, in full.
“achieving balanced accuracies ranging from 85% to 99% from models trained with the Adagarad RDA, AROW, and NHERD online ML algorithms”

Labrèche et al., the flight control team: the mission team. OPS-SAT Spacecraft Autonomy with TensorFlow Lite, Unsupervised Learning, and Online Machine Learning, 2022 IEEE Aerospace Conference (Labrèche and six co-authors, the flight control team). Published 5 Mar 2022 (publication date held by OpenAlex). Read 4 Oct 2026, abstract in full, as held by OpenAlex; the paper itself could not be reached. Read as held by OpenAlex.
“The ability to train models in-flight with data generated on-board without human involvement is an exciting first”

Labrèche et al., the flight control team: the mission team. OPS-SAT Spacecraft Autonomy with TensorFlow Lite, Unsupervised Learning, and Online Machine Learning, 2022 IEEE Aerospace Conference (Labrèche and six co-authors, the flight control team). Published 5 Mar 2022 (publication date held by OpenAlex). Read 4 Oct 2026, abstract in full, as held by OpenAlex; the paper itself could not be reached. Read as held by OpenAlex.
“image classification with Convolutional Neural Network (CNN) model inferences using TensorFlow Lite”

Labrèche et al., the flight control team: the mission team. OPS-SAT Spacecraft Autonomy with TensorFlow Lite, Unsupervised Learning, and Online Machine Learning, 2022 IEEE Aerospace Conference (Labrèche and six co-authors, the flight control team). Published 5 Mar 2022 (publication date held by OpenAlex). Read 4 Oct 2026, abstract in full, as held by OpenAlex; the paper itself could not be reached. Read as held by OpenAlex.
OBJECT_NAME: OPS-SAT; LAUNCH_DATE: 2019-12-18; DECAY_DATE: 2024-05-22

CelesTrak satellite catalogue: a public catalogue. Satellite catalogue record 44878, CelesTrak. Published undated (a live catalogue record; the access date is the date that matters). Read 4 Oct 2026, in full.

Claims to a first

First model trained in space: unchecked. Hedged by its makers ("to our knowledge"). JPL's page on EO-1 speaks of onboard machine learning from 2003 without saying whether anything was trained on board.

Notes

Evidence 2 rests on the paper's abstract, read in full. The paper's body could not be reached.

Changed

Run raised from R1 to R2 after the paper's abstract was read.

This record as data