Onboard autonomy for spacecraft: Junction Brief – November 2026

Acentric Research is produced with AI tools under human review. The authors are not financial professionals, and nothing here is investment advice or a recommendation to buy or sell any security. AI analysis can be wrong; check every claim against its cited source.

This Brief is an initial scan, not a full investigation. The method and scores behind it are in AI × Space in the Landscape Report.

Overview

Most spacecraft do what ground teams tell them. Onboard autonomy moves some decisions to the spacecraft: where it is, which maneuver to make, what to photograph, what to send home. That matters most where messages are slow or ground contacts are few. NASA says that the farther it travels toward the Moon and Mars, the more missions will need to function autonomously. [1]

What stands out is how long parts of this have been flying and how varied the systems are. A JPL system began running on a NASA satellite in 2003 per JPL (the team’s 2006 paper says January 2004; the satellite was retired in 2017), and the authors of NASA’s autoNGC suite report its first flight demonstration, on Starling, in 2025. [2][3] Not all of it is machine learning. JPL says its older system used onboard machine learning and pattern recognition together with planning software. [2] NASA’s descriptions of autoNGC speak of autonomous navigation and maneuver planning and do not use the term. [1][3] This Brief covers the whole field and says which is which where the sources do. The sources opened date from 2006 to 2026, and the flown examples are mostly from 2003 to 2017 and 2025; later developments were not searched.

Current state

An early system in orbit. JPL’s Autonomous Sciencecraft Experiment ran on NASA’s Earth Observing-1 (EO-1) satellite. JPL says the software used onboard continuous planning, task execution, and onboard machine learning and pattern recognition to choose what to send down and to retarget the satellite at science events such as volcanic eruptions. [2] A 2006 conference paper by the team says it had been running since January 2004 and was declared fully validated in May 2004 after all 20 onboard autonomy experiments were complete. [4] JPL’s page lists 67,104 images taken and 1,470 science scenarios executed over the mission, and says EO-1 was decommissioned in spring 2017. [2] NASA’s retirement article calls EO-1 a pathfinder for using artificial intelligence software for streamlined satellite communication. [5]

autoNGC. NASA’s Space Communications and Navigation program describes autoNGC as an onboard software suite that combines navigation, guidance and control functions now split between the spacecraft and the ground, including autonomous maneuver planning for trajectory correction and station-keeping. [1] It says the team demonstrated the software on two Starling CubeSats and on the lunar-orbiting CAPSTONE mission, with a test planned on CAPSTONE 2. [1] The authors’ conference abstract on the 2025 Starling test says the software was uplinked to a spacecraft already in orbit, and that it took GPS measurements, predicted the spacecraft’s state over the next three days and computed station-keeping maneuvers. They report that all three goals were met and that the test raised the software’s readiness to the level of a demonstration in low Earth orbit. [3] Starling is a NASA mission of four CubeSats testing swarm technologies without ground resources. [6]

ESA. ESA’s Intelligent System Initiative aims at what it calls leapfrog advances in system autonomy and a prototype autonomous satellite, with a work plan that includes an AI-based system for failure isolation and recovery on communications constellations and a software environment for intelligent applications. [7] Separately, ESA Discovery funded 12 six-month studies of AI and advanced computing for satellites in a call that opened in October 2021, and ESA says its OPS-SAT satellite had already used AI to process images in flight. [8]

How far it has come. Demonstrated in orbit: an older planning and image-analysis system that ran on EO-1 from 2003, and navigation and maneuver planning in a 2025 flight test. ESA’s work is at the study and building-block stage in the sources opened.

Opportunities

NASA says autoNGC lowers development, test and operating costs, and expects it to enable distributed missions, in-space assembly and low-latency operations. [1] The Starling authors say the autonomy is especially valuable for constellations and other demanding distributed missions, and that the software is available on NASA’s software catalog for other missions. [3] JPL describes the older system as letting a mission study short-lived events and use a fixed downlink better. [2]

Challenges

  • Engineering. The Starling authors say the test highlighted opportunities for further improvement, and NASA is building a next hardware version with more processing and memory. [3][1] Onboard computing is the subject of the space-qualified computing Junction.
  • Capital and scale-up. One of ESA’s planned in-orbit experiments for its initiative was cancelled and is to be replaced by industry-proposed experiments. [7] The Discovery studies were funded for knowledge and skill development rather than specific results. [8]

Data sources

  • NASA mission pages, the NASA Software Catalog and the NASA Technical Reports Server, for flight results. [6][9]
  • ESA programme pages and the esa-star portal, for studies and tenders. [7][10]
  • Conference papers such as the Small Satellite Conference, for flight tests. [3]
  • No registry records which spacecraft run AI onboard or how.
  • No opened source says how many flown systems use machine learning rather than conventional planning, or how mission teams check onboard decisions.

Changelog

  • 2026-11-18: First published.

Sources

  1. NASA Space Communications and Navigation (SCaN) Program, “autoNGC: Autonomous Navigation, Guidance & Control”, program page (agency statement). nasa.gov (accessed 2026-11-18)
  2. NASA Jet Propulsion Laboratory, “Autonomous Sciencecraft Experiment (ASE)”, project page (agency statement). ai.jpl.nasa.gov (accessed 2026-11-18)
  3. Robert Pritchett and others (NASA Goddard Space Flight Center and partners), “Flight Test of the autonomous Navigation, Guidance, and Control (autoNGC) Software on the Starling Swarm”, Small Satellite Conference 2026, abstract (authors’ statement; the presentation file was not read). digitalcommons.usu.edu (accessed 2026-11-18)
  4. Rob Sherwood, Steve Chien, Daniel Tran and others, “Enhancing Science and Automating Operations using Onboard Autonomy”, SpaceOps 2006 conference paper, NASA Jet Propulsion Laboratory and Goddard Space Flight Center. ai.jpl.nasa.gov (accessed 2026-11-18)
  5. Kasha Patel, NASA Goddard Space Flight Center, “A Farewell to EO-1: Celebrating 17 Years of NASA’s ‘Little Earth Satellite That Could’”, NASA, 2017-03-30 (agency statement). science.nasa.gov (accessed 2026-11-18)
  6. NASA, “Starling”, mission page (agency statement). nasa.gov (accessed 2026-11-18)
  7. European Space Agency, “Intelligent System Initiative”, ARTES Advanced Technology programme page, with its work-plan table (agency statement). resilience.esa.int (accessed 2026-11-18)
  8. European Space Agency, Discovery element, “Smarter Satellites: ESA Discovery Accelerates AI in Space” (agency statement; the page is undated in the text read). technology.esa.int (accessed 2026-11-18)
  9. NASA Technical Reports Server. ntrs.nasa.gov (listed as a data source, not opened for this Brief)
  10. European Space Agency, esa-star Publication portal (named on the Intelligent System Initiative page as where tenders are published). esastar-publication.sso.esa.int (listed as a data source, not opened for this Brief)