Artificial intelligence in space

Artificial intelligence in space refers to the use of machine learning, deep learning, autonomous planning systems, and other AI techniques across the space industry. Applications span satellite Earth observation analytics, autonomous spacecraft operations, robotic exploration of other worlds, robotic and AI assistants aboard crewed stations, exoplanet discovery, telescope data analysis, space weather forecasting, debris tracking, and large constellation management. Huge data volumes, limited communication windows with ground stations, and light-time delays that rule out real time human control of distant spacecraft have pushed AI into many parts of modern spaceflight.

Overview

Space missions generate data at a scale that long ago outstripped the capacity of human analysts. A single Sentinel-2 Earth observation satellite collects about 1.5 terabytes of data per day after onboard compression.^[1] The James Webb Space Telescope (JWST) is planned around two four-hour ground contacts a day, each returning at least 28.6 gigabytes of recorded science data.^[2] The Vera C. Rubin Observatory lists about 10 terabytes of data per night and about 10 million transient alerts per night among its key numbers, far more than any human team could vet.^[3] Spacecraft operating beyond Earth orbit also face light delays that make real time joystick control impossible: a command needs roughly 3 to 22 minutes to arrive at a rover on Mars, depending on where the two planets are in their orbits.^[4]

These constraints push mission designers toward onboard autonomy and ground side machine learning pipelines. Commercial Earth observation firms train neural networks on labeled imagery to detect ships, aircraft, vehicles, oil tanks, deforestation, flood extents, and structural damage. Space agencies use AI planners to schedule rover activities, select science targets, avoid hazards during descent and landing, and recover from anomalies. Astronomers use deep learning to classify galaxies, identify gravitational lenses, vet exoplanet candidates, and search for fast radio bursts. Satellite operators use automated systems to predict conjunctions, decide when to maneuver, and characterize unknown objects in orbit.

The number of spacecraft has grown quickly. In its 2025 space environment report, the European Space Agency (ESA) counted about 11,000 active payloads among roughly 40,000 tracked objects,^[5] and ESA's DISCOS database listed about 15,900 functioning satellites as of October 14, 2026.^[6] SpaceX's Starlink satellites, which carry an onboard autonomous collision avoidance system,^[7] performed a combined 207,152 propulsive maneuvers in the six months from December 1, 2025 to May 31, 2026, according to SpaceX's report to the US Federal Communications Commission (FCC).^[8] Projections of future traffic vary. The UK's National Space Operations Centre noted in 2024 that some reports suggest more than 60,000 active satellites by 2030,^[9] and a 2025 Slingshot Aerospace release cited 100,000 expected by 2030.^[10]

History

Work on AI software for spacecraft control dates to the mid-1990s. After NASA announced the New Millennium Program in early 1995, spacecraft engineers at the Jet Propulsion Laboratory (JPL) challenged AI researchers at NASA Ames Research Center and JPL to demonstrate an autonomy architecture within five months, and that work led to the Remote Agent.^[11] On May 17, 1999, the Remote Agent Experiment took command of NASA's Deep Space 1 probe during its cruise. It ran for 29 hours starting that day and for another six hours on May 21, and the team reported that it met all of its experiment objectives.^[12] JPL called it the first artificial intelligence software to command a spacecraft.^[13] During the experiment the team created four simulated failures, including a camera that would not turn off and a thruster stuck in the "off" position, and Remote Agent handled them.^[12] The system combined constraint-based planning and scheduling, robust multi-threaded execution, and model-based diagnosis and reconfiguration.^[11] It was named co-winner of NASA's 1999 Software of the Year award, sharing the honor with Genoa from NASA Glenn.^[13]

The NASA Mars Exploration Rovers Spirit and Opportunity, which landed in January 2004, drove partly with stereo-vision autonomous navigation (AutoNav), which covered about 5 percent of their total driving distance.^[14] Opportunity later gained the AEGIS autonomous targeting system. AEGIS, short for Autonomous Exploration for Gathering Increased Science, was uploaded to Opportunity in the winter of 2009 to 2010 and first used after a drive on March 4, 2010, when the rover analyzed a navigation camera image and pointed its panoramic camera at a rock it had chosen.^[15] The software was later ported to Curiosity: it was uplinked in October 2015 and has been in routine science use with the ChemCam laser spectrometer since May 2016.^[16] AEGIS lets the rover identify rock targets in navigation camera imagery and command ChemCam to analyze them without waiting for a ground command cycle.^[17] In its first year of use on Curiosity, AEGIS chose the preferred kind of target in over 93 percent of cases, compared with roughly 24 percent for blind targeting.^[18][16]

On August 27, 2013, Curiosity drove with autonomous navigation for the first time, onto ground that could not be confirmed safe before the drive began. JPL described this as a first for Curiosity and said the software had been adapted from a similar capability used by Opportunity.^[19] AutoNav saw limited use on these missions: Spirit, Opportunity and Curiosity chose it for under 7 percent of their combined driving, and by April 14, 2022, Curiosity had accumulated 1,116.7 meters of AutoNav driving in total.^[20]

The Perseverance rover, which landed in Jezero Crater on February 18, 2021, brought several major upgrades.^[21] Its Vision Compute Element is a co-processor made up of a RAD750 computer and a Computer Vision Accelerator Card that contains a field programmable gate array (FPGA). It handles stereo processing and visual odometry, and JPL redesigned the mobility software so that images are captured and processed while the rover is driving.^[22] In its first Mars year of operations, AutoNav evaluated 88 percent of the 17.7 kilometers Perseverance traveled.^[23] On October 11, 2026, the mission's 2,006th sol, Perseverance's total distance hit 45.22 kilometers, beyond the 45.16 kilometer off-world driving record set by Opportunity, and NASA said the rover had driven more than 90 percent of its total distance using AutoNav.^[14] The same mission introduced Terrain Relative Navigation during entry, descent, and landing, with a Lander Vision System matching descent imagery against an onboard map of the landing zone.^[21] The rover landed 5 meters from its targeted location, an order of magnitude better than the 60 meter touchdown error requirement.^[21]

The Ingenuity helicopter, which made its first powered flight on Mars on April 19, 2021,^[24] flew under autonomous onboard control on a Qualcomm Snapdragon 801 processor, using a downward-pointing camera for feature tracking, a cellphone-grade inertial measurement unit, and a laser altimeter.^[25] Its flight software used JPL's open source F Prime framework on a Linux operating system,^[26][25] and an Ingenuity engineer said it was the first time Linux would fly on Mars.^[25] Ingenuity was designed to make up to five experimental flights over 30 days; it made 72 flights, the last on January 18, 2024.^[27]

ESA's Phi-Sat-1 launched on a Vega rocket on September 3, 2020 as part of the FSSCat mission.^[28][29] ESA describes it as the first artificial intelligence carried on a European Earth observation mission.^[29] Its deep neural network, running on an Intel Movidius Myriad 2 vision processing unit, sorted hyperspectral images into cloudy and non-cloudy data so that cloudy images did not need to be downlinked.^[28][30] Ubotica, which supplied the AI technology, announced the first successful in-flight inference on September 28, 2020.^[28]

On October 14, 2024, NASA's Europa Clipper launched on a Falcon Heavy from Kennedy Space Center, beginning a 1.8 billion mile cruise to Jupiter. After it begins orbiting Jupiter in April 2030, it will fly past Europa 49 times.^[31] The spacecraft has a dry mass of 7,145 pounds (3,241 kilograms).^[32] A 2018 JPL paper said the mission envisions limited onboard autonomy that can restore the spacecraft's state quickly enough to resume a flyby's observation plan, instead of entering safe mode and waiting for operators.^[33] JPL's MEXEC software was first created as a prototype demonstration for the project, aimed at radiation-induced resets during flybys.^[34]

Earth observation analytics

Earth observation has been the most commercially active area for AI in space. Private operators run constellations of optical and synthetic aperture radar (SAR) satellites and use machine learning to turn pixels into structured information. The table below summarizes the major players.

Larger software platforms aggregate these feeds. Microsoft's Planetary Computer offers more than 140 open datasets and over 50 petabytes of Earth observation and environmental data, according to a 2025 Microsoft conference talk.^[48] Microsoft added Planetary Computer Pro, an enterprise service for an organization's own geospatial data that integrates with Microsoft Fabric, Azure AI Foundry and Power BI;^[49] it was announced in preview in 2025 and reached general availability in June 2026.^[50] Google Earth Engine offers built in ML APIs for supervised and unsupervised classification and regression,^[51] and through ee.Model.fromVertexAi users can call TensorFlow, PyTorch or AutoML models hosted on Vertex AI.^[52]

NVIDIA's Earth-2 shows how the same data is used to train AI weather and climate models. NVIDIA announced Earth-2 in November 2021 as a planned digital twin to simulate and predict climate change,^[53] launched it as a climate digital twin cloud platform in March 2024,^[54] and in January 2026 announced three new open models, Earth-2 Medium Range, Nowcasting and Global Data Assimilation, releasing the first two at once and saying the third would follow later in the year, to join the existing open CorrDiff and FourCastNet3 models.^[55] NVIDIA claims CorrDiff generates images at 12.5 times higher resolution than current numerical models, 1,000 times faster and 3,000 times more energy efficiently.^[54]

Commercial applications cluster in a few areas. Precision agriculture uses ML over multispectral and SAR imagery to compute NDVI time series and crop yield forecasts. Insurance and reinsurance carriers buy automated flood, wildfire, and hail damage assessments. Defense and intelligence customers consume vessel tracking, aircraft activity reports, and infrastructure change alerts. Disaster response agencies use SAR derived flood masks because radar penetrates cloud cover, which is usually thick over flooded areas. After the February 2023 earthquakes in Türkiye and Syria, Microsoft's AI for Good Lab used satellite imagery from the first three days to produce building-level damage estimates for four cities, estimating that 3,849 buildings were damaged or destroyed.^[56]

Onboard processing is spreading as well. In July 2025 NASA and JPL tested "Dynamic Targeting" on the CogniSAT-6 cubesat: for the first time, NASA said, a satellite looked ahead along its orbit, analyzed imagery with onboard AI and decided where to point its instrument, all in less than 90 seconds.^[57]

Spacecraft autonomy

The Mars rovers remain the most visible demonstration of autonomy in deep space. Beyond AutoNav and AEGIS already described, Perseverance's Enhanced AutoNav (ENav) software ranks a set of candidate paths for the rover and then uses the Approximate Clearance Evaluation (ACE) algorithm to check whether the top-ranked paths are safe.^[58] Before Perseverance, the largest total autonomously evaluated driving distance for a Mars rover was 2.4 kilometers, set by Opportunity over its 14-year mission.^[23]

Mission planning has moved partly onto the rover too. Curiosity's scientists and engineers plan activities on the ground with MSLICE (Mars Science Laboratory InterfaCE), a tool for visualizing returned data and collaboratively planning the rover's work.^[59] Perseverance carries an onboard scheduler, the On Board Planner (OBP), which entered primary operations on October 5, 2023^[60] and lets the rover add, extend or delay activities when tasks finish earlier or later than planned. By January 29, 2025 it had operated for 429 sols and executed more than 7,800 activities.^[61]

Perseverance has set several autonomy records. On the relatively flat floor of Jezero Crater it made a single-day drive of 347.7 meters and its longest drive without human review, 699.9 meters.^[62][23] Between June 26 and July 31, 2023, it crossed the boulder field nicknamed Snowdrift Peak in six autonomous drive sols, about a third of the time JPL estimated other NASA rovers would have taken.^[62] In a one-month Rapid Traverse campaign that ended on April 14, 2022, more than 94 percent of its driving used full onboard autonomy, and it logged 4,802 meters of AutoNav driving.^[20]

On December 8 and 10, 2025 (sols 1,707 and 1,709), Perseverance made what JPL called the first drives on another world planned by artificial intelligence, using generative AI to create the waypoints that human rover planners normally choose. The rover drove 210 meters and 246 meters along waypoints that vision language models generated from high-resolution HiRISE orbital imagery and terrain-slope data from digital elevation models. JPL's Rover Operations Center led the work in collaboration with Anthropic, using its Claude models, and before uplink the drive commands were run through JPL's digital twin of the rover, where over 500,000 telemetry variables were verified.^[63] In February 2026 JPL began using Mars Global Localization in regular operations, first on February 2 and again on February 16. The software lets Perseverance determine its position within about 25 centimeters in about two minutes by matching its own images to orbital maps, a task that previously required mapping experts on Earth and could take a day or more. It runs on a commercial processor in the Helicopter Base Station that once relayed communications with Ingenuity.^[64]

ESA's Phi-Sat-1 set the template for AI on orbit in Earth observation.^[30] Its successor, Phi-Sat-2, launched on a Falcon 9 rideshare (Transporter-11) on August 16, 2024.^[65] Its onboard apps at launch covered cloud detection, street map generation, maritime vessel detection, and onboard image compression and reconstruction, and ESA said marine anomaly detection and wildfire detection apps would be uploaded once it was in orbit.^[65] The apps run on Ubotica's CogniSAT platform.^[66] Phi-Sat-2 began its science phase in July 2025.^[67]

NASA's Autonomous Sciencecraft Experiment flew on the Earth Observing-1 satellite from 2003. It used onboard machine learning and pattern recognition to detect events such as volcanic eruptions, flooding, ice breakup and cloud cover change, and to retarget the satellite without ground intervention.^[68][69] It met its full success criteria in May 2004,^[69] and EO-1 was decommissioned in spring 2017.^[68]

AEGIS has also been deployed to Perseverance's SuperCam instrument, which first used it on sol 442, May 18, 2022. As of May 11, 2026, AEGIS had selected 293 Perseverance targets, and in one campaign it was used in 45 of 48 uplink plans.^[16] The system has the same role as on Curiosity: pick rocks for the laser spectrometer without waiting for the next uplink.

AI and robots aboard the International Space Station

The International Space Station has hosted free-flying robots since the SPHERES units arrived in 2006,^[70] and a series of robotic and AI assistants since Robonaut 2 in 2011. Most are teleoperated, commanded from the ground, or carry out pre-planned tasks under supervision. Their AI elements range from cloud-based speech recognition to experimental machine learning control and language models run on station computers. Several projects make competing "first" claims, which are each developer's own wording.

Robonaut 2

Robonaut 2 (R2) was developed jointly by NASA and General Motors.^[84] NASA delivered it to the station on the space shuttle Discovery's final flight in February 2011, and the crew first powered it up on August 22, 2011; it was the first humanoid robot to fly in space.^[71] R2 could press buttons, flip switches and turn knobs, and crew members could teleoperate it. A pair of legs launched to the station in April 2014. After problems traced in part to a missing ground path in its wiring, R2 returned to Earth for repairs on May 5, 2018, aboard SpaceX's fourteenth Dragon cargo flight.^[72] It has not flown since; from October 24, 2024, NASA lent it to the Smithsonian's National Air and Space Museum for display at the Udvar-Hazy Center.^[190] A NASA paper from its early years on the station described operations in which the crew or ground controllers ran semi-autonomous scripts, with teleoperation, shared control and supervised autonomy among the planned modes.^[73] Johnson Space Center's engineering directorate built on its Robonaut 2 experience when it designed the full-body Valkyrie humanoid for the 2013 DARPA Robotics Challenge Trials.^[85]

CIMON

CIMON (Crew Interactive MObile companioN) is a free-flying, voice-controlled assistant built by Airbus for the German Space Agency at DLR, with speech recognition and language understanding based on IBM's Watson technology. CIMON-1 launched on June 29, 2018, aboard a Dragon capsule on the SpaceX CRS-15 mission.^[75] Its first session with ESA astronaut Alexander Gerst began on November 15, 2018 and lasted 90 minutes. Speech was converted to text and processed by Watson in the IBM Cloud in Frankfurt, and the replies were converted back to speech and sent to the station; Airbus called it the first AI-based astronaut assistant.^[86] CIMON-1 returned to Earth when the SpaceX CRS-18 Dragon undocked on August 27, 2019, after 14 months on the station.^[87]

CIMON-2 launched on December 5, 2019, on SpaceX CRS-19. It could use the IBM Watson Tone Analyzer to evaluate the emotions of astronauts. DLR noted that CIMON had no self-learning capabilities and that the AI it used for autonomous navigation came from Airbus.^[76] It was deployed in February 2020, and in its debut session ESA astronaut Luca Parmitano asked it to fly to the Biolab facility.^[88] Swedish ESA project astronaut Marcus Wandt used it during the Axiom-3 mission in 2024, when CIMON guided him through a 3D kinetic gas model experiment and recorded it.^[89] On September 19, 2024, NASA astronauts Mike Barratt and Jeanette Epps tested CIMON's scientific capabilities using voice prompts.^[90]

In the ICHIBAN experiment, carried out on July 29, 2025 and announced by DLR two days later, CIMON and JAXA's Int-Ball2 communicated with each other for the first time. JAXA astronaut Takuya Onishi spoke to CIMON in the Columbus module, and Int-Ball2, controlled through CIMON, searched for and photographed objects hidden in the Kibo module. CIMON's AI by then used IBM's watsonx technology.^[91]

Astrobee

NASA's Astrobee system, built at Ames Research Center, consists of three free-flying robots named Bumble, Honey and Queen. Bumble and Honey launched on April 17, 2019, on Northrop Grumman's 11th commercial resupply mission, and Queen followed on July 25, 2019, on SpaceX's 18th.^[77] On June 14, 2019, Bumble became the first Astrobee to fly under its own power in space. Its navigation system compares what its camera sees with a map of the station's interior.^[92] Astrobee replaced the SPHERES free-flyers that had supported guest science on the station since 2006. In NASA's account, it has carried out more than 165 real-time operations, and nearly 80 percent of console time has not required the crew.^[70] In April 2022 the Integrated System for Autonomous and Adaptive Caretaking (ISAAC) project had two Astrobees, Queen and Bumble, work independently in separate modules for the first time.^[93][77]

Astrobee became a test bed for machine learning control in 2025. On May 27, 2025, a US Naval Research Laboratory team controlled Bumble with a reinforcement learning policy trained with proximal policy optimization in NVIDIA Omniverse; NRL called it the first reinforcement learning control of a free-flyer in space.^[94] On the same day, Honey autonomously docked with Bumble using Obruta's AI-based monocular vision software.^[95] Stanford University researchers reported in December 2025 that a machine learning model trained on thousands of earlier path solutions gave Astrobee's trajectory optimizer a "warm start" that made planning 50 to 60 percent faster, while the optimizer still enforced the safety constraints. Stanford described it as the first time AI had been used to help control a robot on the station.^[96] NASA awarded Arkisys a Space Act Agreement in September 2025 to sustain and maintain the robots,^[77] and in January 2026 Arkisys announced that Astrobee had returned to operational status.^[97] NASA expects Astrobee to keep operating until the station is retired.^[70]

Int-Ball and Int-Ball2

JAXA's Int-Ball camera drone reached the station on a Dragon launched June 4, 2017, and JAXA released its first images on July 14, 2017. It moves autonomously and records still and moving images under remote control from JAXA's Tsukuba Space Center. JAXA's goal was to cut the time astronauts spend taking photographs, then about 10 percent of their working hours, to zero.^[74] Int-Ball2 launched in June 2023, and JAXA astronaut Satoshi Furukawa carried out its initial checkout that October; it flies under the control of the JAXA flight control team. On November 7, 2023, it docked with and released from its charging port automatically for the first time.^[189] On September 20, 2024, during the final run of the Kibo Robot Programming Challenge, Int-Ball2 and Astrobee flew together for the first time.^[82] JAXA says Int-Ball2 can also serve as a technology demonstration platform on which outside users run their own programs, and in March 2025 it released a ROS/Gazebo simulator for that purpose.^[83]

Other robots and onboard computing

Russia's Skybot F-850, a humanoid robot about 1.8 meters tall that evolved from the FEDOR series of rescue robots, flew to the station on Soyuz MS-14, which launched on August 22, 2019 and arrived on August 26 after an aborted first docking attempt. Cosmonaut Alexander Skvortsov wore virtual reality headgear to control it, and it connected cables and used spacewalk tools before returning to Earth on September 6.^[78] Japanese startup GITAI's S1 robotic arm, which flew on SpaceX's CRS-23 mission, assembled structures and panels and worked switches and cables on its own in the Nanoracks Bishop airlock from October 13 to 16, 2021; the switch and cable tasks were also run by teleoperation from Nanoracks mission control in Houston.^[187][81] GITAI said in March 2024 that its 1.5-meter dual-arm S2 system had finished every planned servicing, assembly and manufacturing task outside the Bishop airlock.^[188]

Hewlett Packard Enterprise's Spaceborne Computer-2, launched in February 2021, has run experiments in fields such as DNA sequencing and image processing.^[79] One, developed by NASA, Microsoft and HPE, used a Glove Analyzer model trained with Azure Custom Vision to look for damage in photos of astronauts' gloves; Microsoft said it was in a trial stage and not used to make crucial safety decisions.^[98] On Axiom Space's Ax-1 mission in 2022, a machine learning model running on an Amazon Web Services Snowcone device analyzed photos taken aboard the station and returned results within three seconds.^[99] In August 2024 Booz Allen Hamilton said it had run a generative AI large language model on Spaceborne Computer-2, showing that techniques such as retrieval-augmented generation could be used there, and that the model was believed to be the first LLM deployed in space,^[100][79] and in April 2025 Booz Allen and Meta began deploying a fine-tuned version of Llama 3.2 on the same computer, a project the two companies called "Space Llama."^[101][102]

Exoplanet discovery and astronomy

The Kepler space telescope launched in 2009 and, after its K2 extended mission, was retired in 2018,^[103] and the Transiting Exoplanet Survey Satellite (TESS) has been operating since its launch on April 18, 2018.^[104] Each one produced light curves for hundreds of thousands of stars, and the false positive rate for transit candidates is high. Machine learning has become a standard tool for sorting through them.

In a 2018 paper in The Astronomical Journal, Christopher Shallue and Andrew Vanderburg trained a one dimensional convolutional neural network on phase folded Kepler light curves.^[105] The model found Kepler-90i, an eighth planet around Kepler-90, and the authors statistically validated a new planet in the Kepler-80 system, which forms a five-planet resonant chain.^[105]

ExoMiner, whose project lead is Hamed Valizadegan of the Universities Space Research Association at NASA Ames, took the approach further.^[106] The 2021 paper, published in The Astrophysical Journal in 2022, validated 301 new planets from the Kepler archive using a deep neural network built on the same diagnostics that human vetters examine.^[107][106] At a fixed precision of 99 percent, the model recovered 93.6 percent of the exoplanets in its test set.^[107] A 2023 follow-up validated 69 more Kepler planets,^[108] and in 2025 an extended version, ExoMiner++, applied to TESS data identified 7,330 signals as planet candidates.^[109] These planets are statistically validated by the classifier rather than confirmed through independent observations.^[107]

Deep learning also contributes to JWST analysis. The Morpheus framework, developed at UC Santa Cruz by Ryan Hausen and Brant Robertson, performs pixel-level classification of astronomical images,^[110] and its authors used it for what they called the first AI/ML analysis of JWST images, classifying distant disk galaxies.^[111] NVIDIA said in 2022 that Robertson's team would use Morpheus to classify findings from the JWST Advanced Deep Extragalactic Survey (JADES),^[112] and in 2026 Robertson told NVIDIA that "even teams of experts would take years to do what now needs to happen in days."^[113]

The Vera C. Rubin Observatory in Chile took its first on-sky engineering data with the LSST Camera on April 15, 2025,^[114] and released its first public images on June 23, 2025.^[115] Each image is meant to travel from the mountain to the SLAC National Accelerator Laboratory in under seven seconds.^[116] Rubin's alert system, launched in February 2026, compares new images with reference templates and can issue a public alert within about two minutes, with up to 7 million alerts per night by SLAC's count (Rubin's key-numbers page gives about 10 million); community brokers then use machine learning algorithms to filter, sort and classify the alerts.^[117] The official LSST Science Pipelines are free software available on GitHub.^[118] A January 2026 white paper from the Rubin LSST Dark Energy Science Collaboration on AI and ML opportunities notes that Rubin's catalog is expected to exceed 20 billion galaxies, with about 10 million alerts per night, and calls the detection of novel phenomena in that data an unprecedented challenge.^[119]

The Euclid space telescope, launched by ESA on a Falcon 9 on July 1, 2023, started its scientific survey on February 14, 2024. Over the survey it is designed to measure the shapes of about 1.5 billion galaxies for weak lensing cosmology.^[120] ESA's first Quick Data Release, Q1, appeared on March 19, 2025, covering 63 square degrees and 26 million galaxies; the Zoobot AI model classified more than 380,000 of them, and a combination of AI and citizen science produced a first catalog of 500 galaxy-galaxy strong lens candidates. ESA expects some 7,000 candidates in the major cosmology data release and on the order of 100,000 by the end of the mission.^[121] The first full data release has since been split in two: as of a June 15, 2026 update to ESA's DR1 timeline page, a DR1-Foundation release covering about 1,900 square degrees is due in November 2026, with the complete DR1, including the higher-level galaxy clustering and weak lensing products, following in mid-2027.^[122]

Machine learning has also driven a series of results in gravitational wave astronomy and radio transient detection. Daniel George and E. A. Huerta showed in 2017 and 2018 that deep convolutional neural networks could detect LIGO binary black hole merger signals and estimate their parameters in real time, with performance similar to matched filtering at much greater speed.^[123] The Gravity Spy project combines machine learning with citizen science to classify glitches in LIGO data.^[124] Liam Connor and Joeri van Leeuwen's 2018 paper in The Astronomical Journal applied a tree-like deep neural network to classify fast radio burst (FRB) candidates, training it separately on data from the CHIME Pathfinder and from Apertif on the Westerbork telescope.^[125]

The COSMIC system at the Karl G. Jansky Very Large Array is operated by the SETI Institute in collaboration with the University of California, Berkeley and Breakthrough Listen.^[126] Its first VLA Sky Survey results paper filtered candidate signals with statistical tests and a series of logical steps, and its authors wrote that such post-processing, done by humans or by machine learning, will become necessary as data collection grows.^[127] Breakthrough Listen researchers reported in Nature Astronomy in 2023 that a deep learning search of more than 480 hours of Green Bank Telescope data on 820 nearby stars cut the number of candidate signals by about two orders of magnitude compared with earlier analyses of the same data and found eight new signals of interest.^[128] Re-observations of those targets have so far not detected the signals again.^[128]

Space weather

The Sun drives a long list of operational risks: geomagnetic storms that knock out HF radio and damage power grids, energetic particle events that endanger astronauts, and atmospheric drag changes that pull low Earth orbit satellites down faster. A 2025 study in Space Weather by Enrico Camporeale and Thomas Berger verified the solar flare forecasts issued by NOAA's Space Weather Prediction Center (SWPC) from 1998 to 2024. It found that the SWPC model did not outperform simple baselines such as persistence, climatology, naive Bayes and logistic regression, and that its forecasts had severe calibration issues and high false alarm rates.^[129]

The Frontier Development Lab (FDL), now run by FDL.AI, a division of Trillium Technologies, with computing support from Google Cloud and NVIDIA, has produced a long list of heliophysics tools for NASA and other partners.^[130] DAGGER (Deep leArninG Geomagnetic pErtuRbation) predicts geomagnetic disturbances worldwide 30 minutes before they occur, using solar wind observations.^[131][132] The related SHEATH pipeline was paired with DAGGER as SHEATH-DAGGER, which FDL says provides forecasts with uncertainty estimates.^[133] FDL's MEGS-AI project estimates the full extreme ultraviolet spectrum from Solar Dynamics Observatory images.^[133] Separately, Robert Jarolim and colleagues published an instrument-to-instrument translation framework in 2025 that uses deep learning to make observations from different solar instruments consistent,^[134] and the same lead author's SuNeRF model uses neural radiance fields to reconstruct the solar corona in 3D.^[135]

Other teams have compared convolutional networks with vision transformers for classifying solar active regions in Solar Dynamics Observatory magnetogram and continuum images.^[136] A January 2025 arXiv preprint by Sabrina Guastavino and colleagues argues that AI could have predicted all the space weather events associated with the May 2024 superstorm.^[137]

Space situational awareness and debris

The US Space Force's 18th Space Defense Squadron maintains the public satellite catalog.^[138] As of October 14, 2026, ESA's Space Debris Office put the number of objects regularly tracked by space surveillance networks and kept in their catalogs at about 47,080. ESA's MASTER-8 model estimates about 68,450 objects larger than 10 cm, 1.5 million between 1 and 10 cm, and 230 million between 1 mm and 1 cm.^[6] The commercial sector has built its own networks. LeoLabs operates tracking radars in Australia, the Azores, Costa Rica, New Zealand and Texas, uses the Poker Flat radar in Alaska through a partnership with SRI International, and lists newer "Seeker" radars for Arizona and the Indo-Pacific as in development.^[139] A 2024 paper by LeoLabs engineers described XGBoost classifiers that determine an object's type and maneuverability from radar data within 72 hours of association. Mean cross-validated accuracy was about 91 percent for object type and 92 percent for maneuverability, and up to 95 percent for some object types.^[138] LeoLabs also uses AI and machine learning models for maneuver detection, a daily pattern-of-life analysis that gives objects with unusual activity an "interest score," object characterization and radar scheduling.^[140]

Slingshot Aerospace said in October 2025 that it operates 204 sensors in 21 locations across five continents.^[10] In June 2024 Slingshot announced Agatha, an AI system it developed with DARPA that identifies anomalous spacecraft within large satellite constellations and uses inverse reinforcement learning.^[141] In November 2024 NOAA awarded Slingshot the contract for the presentation layer of its Traffic Coordination System for Space (TraCSS), worth $5.3 million over 12 months, with options.^[142]

Privateer, co-founded by Steve Wozniak, chief executive Alex Fielding and chief scientist Moriba Jah, was announced in September 2021 and came out of stealth on March 1, 2022 with Wayfinder, a 3D globe for tracking satellites and space debris.^[143] In May 2024 it acquired the analytics company Orbital Insight and raised $56.5 million, Reuters reported.^[144]

In Europe, the EU Space Surveillance and Tracking (EU SST) framework was set up by an EU decision in 2014 and was brought under the regulation establishing the EU Space Programme in 2021.^[145] Four more member states joined the EU SST Partnership in April 2026, bringing it to 19.^[146] EUSPA says EU SST supports more than 690 satellites and has over 400 registered organisations, providing collision avoidance, analysis of objects re-entering the atmosphere, and assessment of in-orbit break-ups.^[147] ESA's CREAM (Collision Risk Estimation and Automated Mitigation) project, started in 2020, is developing automated collision avoidance; by August 2025 it was in ground system test operations, still ground-based, with in-orbit demonstrations being prepared.^[148]

Constellation operations

SpaceX's Starlink is the largest example of autonomous constellation operations. SpaceX says every one of its satellites carries an onboard, autonomous collision avoidance system that plans avoidance maneuvers when the probability of collision passes a set threshold.^[7] That threshold has tightened over time: greater than 1 in 100,000 in SpaceX's February 2022 description,^[7] 1 in 1,000,000 by the report covering December 2023 to May 2024,^[149] and, in SpaceX's July 2026 report, greater than 3 in 10,000,000, along with additional miss-distance criteria, against an industry standard of 1 in 10,000.^[8] Conjunctions are screened by the US 19th Space Defense Squadron in screening cycles every eight hours,^[8] and SpaceX shares its satellites' predicted ephemerides and covariance data on Space-Track.org.^[7] First-generation Starlink satellites use krypton-fueled ion thrusters, while the V2 Mini satellites use argon-fueled Hall thrusters.^[150]

SpaceX reports its maneuver counts to the FCC twice a year. Selected figures:

Under a Space Act Agreement signed in January 2021, NASA agreed not to maneuver its assets for potential conjunctions with Starlink satellites and to rely on the Starlink satellites' autonomous maneuvering, while allowing for cases such as Starlink anomalies in which NASA assets would have to maneuver.^[152] China told the United Nations that its Tiangong space station had maneuvered to avoid Starlink satellites twice in 2021, in July and October.^[153] In its July 2026 report SpaceX described a "time-domain deconfliction" approach, agreed with some other operators, in which the satellite transiting Starlink's orbits takes maneuver responsibility on long time horizons and SpaceX keeps it closer to the conjunction.^[8]

In January 2026 SpaceX announced Stargaze, a space situational awareness system that draws on nearly 30,000 star trackers across its fleet, which it said detect about 30 million transits a day. SpaceX said Stargaze can assess possible conjunctions on the order of minutes instead of the several hours that is the current standard, and that it would make the conjunction data available to all operators free of charge.^[154]

Eutelsat completed its combination with OneWeb on September 28, 2023.^[155] The Eutelsat OneWeb constellation has over 600 satellites in 12 orbital planes 1,200 km above the Earth, and in September 2026 Airbus announced an order from Eutelsat for 229 more OneWeb satellites.^[156] In May 2025 Ramon.Space won a Eutelsat OneWeb contract to supply onboard digital channelizers, initially 70 flight sets, which the company says can be reprogrammed in orbit.^[157]

Amazon's Project Kuiper, rebranded as Amazon Leo in November 2025,^[158] Telesat Lightspeed, and China's Guowang and SpaceSail constellations will add thousands more satellites. Amazon's 2023 description of its safety approach covers active propulsion, sharing ephemeris through Space-Track and coordination with NASA.^[159]

NASA and ESA AI programs

US federal agencies inventory their AI use cases under Executive Order 13960 of December 3, 2020, "Promoting the Use of Trustworthy Artificial Intelligence in the Federal Government," which also sets out principles for the trustworthy use of AI in government, an approach often discussed as responsible AI.^[160] NASA's fiscal 2024 inventory listed 18 use cases, among them AEGIS, Enhanced AutoNav for Perseverance, Terrain Relative Navigation, the Mars 2020 Onboard Planner and the SensorWeb system for volcanoes, floods and wildfires.^[161] The fiscal 2025 inventory listed 425, including ExoMiner, the TAPAS tool for finding anomalies in Deep Space Network telemetry, AI-assisted spacecraft fault detection, wildfire detection, and ChatGSFC, a chatbot for Goddard Space Flight Center staff.^[162]

The Frontier Development Lab, already discussed, centers each year on an eight to nine week research sprint by interdisciplinary four-person teams of PhD and postdoc level researchers. Its Moon for Good team produced the first map of metallic resources distributed on the lunar surface,^[130] and astronaut health was a 2019 theme.^[163]

JPL's Artificial Intelligence Group built the Autonomous Sciencecraft software for Earth Observing-1, the CASPER onboard planner, and AEGIS.^{[68][164][16]} CASPER flew on Earth Observing-1 for more than a dozen years and on the IPEX cubesat for over a year.^[164] The group's MEXEC planning and execution software has been used for operations on the ASTERIA cubesat, and its timeline module is maintained by the Mars 2020 mission as part of the On Board Planner.^[34] Remote Agent was a joint effort with NASA Ames, where much of its code was written.^[12][13]

FDL has a European sister lab run with ESA.^[130] ESA's Phi-lab works on disruptive innovations for Earth observation, and through the InCubed programme it offers financial, business and technical support to innovations with strong commercial potential.^[165] ESA's OPS-SAT in-orbit laboratory, launched on December 18, 2019, ran until its re-entry in May 2024; 134 teams from 26 countries ran more than 284 experiments on it.^[166]

Defense space

The Space Development Agency, established on March 12, 2019,^[167] became part of the US Space Force in October 2022.^[168] It is building the Proliferated Warfighter Space Architecture, a constellation launched in tranches for missile tracking and tactical communications; Tranche 0 had 27 satellites, and Tranche 1 is planned at 154 operational spacecraft.^[169]

SDA's Prototype On-orbit Experimental Testbed (POET) hosted early demonstrations of AI-enabled edge processing by Scientific Systems Company on Loft Orbital's YAM-3 satellite, which launched on June 30, 2021. The first application was a machine learning cloud detection tool, and the company announced the demonstrations in February 2022.^[170] DARPA's Hallmark program, which ran from 2015 to 2020, included indications and warnings work that used machine learning to detect threats.^[171] RAND published two 2024 reports on AI and machine learning for space domain awareness, one on the impact on mission effectiveness and one on the development of two AI case studies.^[172][173] In April 2025 AFWERX selected Slingshot for its Rapid Analysis of Photometric Tracks for space Object identification and behavior Recognition (RAPTOR) effort, which uses machine learning to track and identify in-space activity.^[174]

Concerns and limitations

AI in space has limits that mission planners take seriously. Spacecraft hardware lags well behind terrestrial processors because radiation hardened chips run on older process nodes and have far less memory. Ingenuity flew a Snapdragon 801, a mobile chip Qualcomm introduced in 2014,^[175][176] while Perseverance's main computer is a radiation-hardened BAE RAD750 with PowerPC 750 architecture^[177] and many ESA missions, including BepiColombo, Gaia and the Sentinels, use LEON processors.^[178] That gap limits which models can run on board.

Trust in onboard decisions is a separate question. Starlink's collision avoidance is a case where onboard software, not a human, makes the maneuver decision; SpaceX describes it as an autonomous system rather than as machine learning.^[7]

Commercial imagery and the AI built on it also raise policy questions. A 2024 Nature study that applied deep learning to satellite imagery found that 72 to 76 percent of the world's industrial fishing vessels are not publicly tracked.^[179] Access to commercial imagery can also be cut for policy reasons: in March 2025 the US government temporarily suspended Ukrainian accounts on the Global Enhanced GEOINT Delivery platform, which Maxar operated,^[180] and users began receiving notices that access had been restored a few days later.^[181]

False positive and calibration issues recur. Operational solar flare forecasts have shown severe calibration problems and high false alarm rates.^[129] Technosignature searches must contend with radio interference: the BLC1 signal of interest, found toward Proxima Centauri with the turboSETI Doppler search code at the Parkes telescope, turned out to be an intermodulation product of local interference.^[182][183] The planets that exoplanet classifiers validate are statistical results rather than independent confirmations.^[107]

The compute cost of training large Earth observation foundation models is also part of the conversation. IBM and NASA released a new version of their Prithvi geospatial model in December 2024,^[184] IBM and ESA open-sourced TerraMind in April 2025,^[185] and Google DeepMind introduced AlphaEarth Foundations in July 2025.^[186] The questions about compute footprint that apply to text large language models apply to these models as well.

See also

• Space ChatGPT Plugins
• Space-based data centers
• Machine learning
• Deep learning
• Convolutional neural network
• Edge computing
• Humanoid robot
• NASA
• NVIDIA
• Microsoft

References

1. ^Airbus, "Airbus-built Sentinel-2C satellite successfully launched," September 5, 2024. airbus.com/...l-2c-satellite-successfully-launched
2. ^Space Telescope Science Institute, "JWST Solid State Recorder," JWST User Documentation. jwst-docs.stsci.edu/...jwst-solid-state-recorder
3. ^Rubin Observatory, "Key Numbers," Rubin 101 for scientists. rubinobservatory.org/...key-numbers
4. ^Pashai, P., "Speaking Percy's Language: How to Talk so Perseverance Will Listen," NASA Science, September 14, 2021. science.nasa.gov/...lk-so-perseverance-will-listen
5. ^ESA, "ESA Space Environment Report 2025," April 1, 2025. esa.int/...ESA_Space_Environment_Report_2025
6. ^1 ^2ESA Space Debris Office, "Space debris by the numbers" (DISCOSweb), data as of October 14, 2026. sdup.esoc.esa.int/...bythenumbers
7. ^1 ^2 ^3 ^4 ^5SpaceX, "SpaceX's Approach to Space Sustainability and Safety," February 22, 2022. spacex.com/updates
8. ^1 ^2 ^3 ^4 ^5Space Exploration Holdings (SpaceX), "SpaceX Gen 1 and Gen 2 Semi-Annual Reports" (letter to the Federal Communications Commission, reporting period December 1, 2025 to May 31, 2026), July 1, 2026. api-prod.fcc.gov/...bff0244acfb947505791bf252f851ce6
9. ^National Space Operations Centre, UK Space Agency and Ministry of Defence, "The future space environment," GOV.UK, May 16, 2024. gov.uk/...the-future-space-environment
10. ^1 ^2Slingshot Aerospace, "Slingshot Aerospace Advances UK Space Agency's Satellite Tracking Amid Orbital Congestion," October 28, 2025. slingshot.space/...racking-amid-orbital-congestion
11. ^1 ^2Muscettola, N., Nayak, P. P., Pell, B. and Williams, B. C., "Remote Agent: To Boldly Go Where No AI System Has Gone Before," Artificial Intelligence 103(1-2), August 1998. doi.org/...S0004-3702(98)00068-X
12. ^1 ^2 ^3NASA Jet Propulsion Laboratory, "Remote Agent Experiment Meets All Objectives," June 7, 1999. jpl.nasa.gov/...nt-experiment-meets-all-objectives
13. ^1 ^2 ^3NASA Jet Propulsion Laboratory, "Futuristic Software Demonstrated On Deep Space 1 Wins NASA Award," August 6, 1999. jpl.nasa.gov/...ed-on-deep-space-1-wins-nasa-award
14. ^1 ^2NASA Science Editorial Team, "NASA's Perseverance Sets Off-World Record for Total Distance Driven," NASA Science, October 12, 2026. science.nasa.gov/...cord-for-total-distance-driven
15. ^NASA Jet Propulsion Laboratory, "NASA Mars Rover Getting Smarter as It Gets Older," March 23, 2010. jpl.nasa.gov/...r-getting-smarter-as-it-gets-older
16. ^1 ^2 ^3 ^4JPL Artificial Intelligence Group, "AEGIS: Autonomous Exploration for Gathering Increased Science," NASA Jet Propulsion Laboratory. ai.jpl.nasa.gov/...aegis
17. ^NASA Jet Propulsion Laboratory, "NASA Mars Rover Can Choose Laser Targets on Its Own," July 21, 2016. jpl.nasa.gov/...an-choose-laser-targets-on-its-own
18. ^Francis, R. et al., "AEGIS autonomous targeting for ChemCam on Mars Science Laboratory: Deployment and results of initial science team use," Science Robotics 2(7), June 2017. doi.org/...scirobotics.aan4582
19. ^NASA Jet Propulsion Laboratory, "NASA's Mars Curiosity Debuts Autonomous Navigation," August 27, 2013. jpl.nasa.gov/...osity-debuts-autonomous-navigation
20. ^1 ^2Rankin, A., Del Sesto, T., Hwang, P., Justice, H., Maimone, M., Verma, V. and Graser, E., "Perseverance Rapid Traverse Campaign," IEEE Aerospace Conference, 2023. robotics.jpl.nasa.gov/...2023-rapid-traverse.pdf
21. ^1 ^2 ^3Johnson, A. E. et al., "Mars 2020 Lander Vision System Flight Performance," AIAA SciTech 2022 Forum, January 2022. ntrs.nasa.gov/...20230006986
22. ^JPL Robotics, "Mobility" (Mars 2020 rover), NASA Jet Propulsion Laboratory. robotics.jpl.nasa.gov/...m2020mobility
23. ^1 ^2 ^3Verma, V. et al., "Autonomous robotics is driving Perseverance rover's progress on Mars," Science Robotics 8(80), July 12, 2023. doi.org/...scirobotics.adi3099
24. ^NASA Jet Propulsion Laboratory, "NASA's Ingenuity Mars Helicopter Succeeds in Historic First Flight," April 19, 2021. jpl.nasa.gov/...-succeeds-in-historic-first-flight
25. ^1 ^2 ^3Ackerman, E., "How NASA Designed a Helicopter That Could Fly Autonomously on Mars," IEEE Spectrum, February 17, 2021. spectrum.ieee.org/...r-rover-fly-autonomously-mars
26. ^NASA Jet Propulsion Laboratory, "Meet the Open-Source Software Powering NASA's Ingenuity Mars Helicopter," July 8, 2021. jpl.nasa.gov/...ng-nasas-ingenuity-mars-helicopter
27. ^NASA Jet Propulsion Laboratory, "After Three Years on Mars, NASA's Ingenuity Helicopter Mission Ends," January 25, 2024. jpl.nasa.gov/...-ingenuity-helicopter-mission-ends
28. ^1 ^2 ^3Ubotica Technologies, "Satellite Successfully Applies AI to Process Earth Observation Imagery In-flight in Historic First for Space," September 28, 2020. ubotica.com/...-flight-in-historic-first-for-space
29. ^1 ^2ESA, "Artificial Intelligence for Earth observation." esa.int/...cial_Intelligence_for_Earth_observation
30. ^1 ^2Giuffrida, G. et al., "The Φ-Sat-1 Mission: The First On-Board Deep Neural Network Demonstrator for Satellite Earth Observation," IEEE Transactions on Geoscience and Remote Sensing 60, 2022. doi.org/...TGRS.2021.3125567
31. ^NASA Jet Propulsion Laboratory, "Liftoff! NASA's Europa Clipper Sails Toward Ocean Moon of Jupiter," October 14, 2024. jpl.nasa.gov/...sails-toward-ocean-moon-of-jupiter
32. ^NASA Science, "Europa Clipper and People." science.nasa.gov/...europa-clipper-and-people
33. ^Schaffer, S., Chien, S. and Ferguson, E., "Validation of Fault-Tolerant Plans for Europa Clipper," i-SAIRAS 2018. ai.jpl.nasa.gov/...schaffer-isairas2018-clipper.pdf
34. ^1 ^2JPL Artificial Intelligence Group, "MEXEC," NASA Jet Propulsion Laboratory. ai.jpl.nasa.gov/...mexec
35. ^Planet Labs PBC, "Planet Announces Planet Insights Platform with Powerful Tooling for Unlocking Earth Observation Data," April 9, 2024. investors.planet.com/...default
36. ^Planet Labs PBC, "Planet Launches Two Additional High-Resolution Pelican Satellites," August 26, 2025. investors.planet.com/...default
37. ^Planet Labs PBC, "Planet Successfully Runs AI in Space," April 7, 2026. investors.planet.com/...cessfully-Runs-AI-in-Space
38. ^Vantor, "Vantor Rebrands from Maxar Intelligence, Unveils AI-Powered Platform," October 1, 2025. vantor.com/...lligence-unveils-ai-powered-platform
39. ^Maxar, "Open Data Response to the Earthquakes in Turkey and Syria," Maxar Blog, February 7, 2023 (archived). web.archive.org/...earthquakes-in-turkey-and-syria
40. ^Capella Space, "How AI Supercharges SAR Intelligence," July 29, 2025. capellaspace.com/...-supercharges-sar-intelligence
41. ^IonQ, "IonQ Completes Acquisition of Capella Space, Advancing Vision for Space-Based Quantum Communications," July 15, 2025. ionq.com/...space-advancing-vision-for-space-based
42. ^ICEYE, "ICEYE unveils machine learning-powered Flood Rapid Impact Product to revolutionize response," July 14, 2025. iceye.com/...ml-powered-flood-rapid-impact-product
43. ^Microwave Journal, "ICEYE and SATIM Announce Detect & Classify Product for AI-Powered Analysis of SAR Imagery," September 3, 2025. microwavejournal.com/...ed-analysis-of-sar-imagery
44. ^BlackSky, "BlackSky Spectra nominated for Via Satellite's 2023 Satellite Technology of the Year Award," March 6, 2024. blacksky.com/...llite-technology-of-the-year-award
45. ^Satellogic, "Satellogic open-source release: A large dataset of high-resolution imagery for AI model training," May 1, 2024. satellogic.com/...on-imagery-for-ai-model-training
46. ^Satellogic, "Pushing Intelligence to the Edge: Satellogic's Vision for AI-Powered Earth Observation," March 20, 2025. satellogic.com/...for-ai-powered-earth-observation
47. ^Umbra, "Umbra Releases Over $4 Million in Free SAR Data," March 1, 2024. umbra.space/...ses-over-4-million-in-free-sar-data
48. ^Microsoft, "Microsoft's Planetary Computer: Building a Planetary-Scale Data Platform," FOSS4G North America 2025, November 4, 2025. talks.osgeo.org/...GW89SP
49. ^Phillips, D., "Microsoft Planetary Computer Pro: Unlocking AI-powered geospatial insights for enterprises across industries," Microsoft Azure Blog, June 17, 2025. azure.microsoft.com/...terprises-across-industries
50. ^Microsoft Learn, "New Features in Microsoft Planetary Computer Pro." learn.microsoft.com/...whats-new
51. ^Google, "Machine Learning in Earth Engine," Google Earth Engine guides. developers.google.com/...machine-learning
52. ^Google, "Predictions from Hosted Models," Google Earth Engine guides. developers.google.com/...ee-vertex-overview
53. ^NVIDIA, "GTC Wrap-Up: NVIDIA CEO Outlines Vision for Accelerated Computing, Data Center Architecture, AI, Robotics, Omniverse Avatars and Digital Twins in Keynote," NVIDIA Blog, November 9, 2021. blogs.nvidia.com/...i-omniverse-avatars-robots-gtc
54. ^1 ^2NVIDIA, "NVIDIA Announces Earth Climate Digital Twin," March 18, 2024. nvidianews.nvidia.com/...arth-climate-digital-twin
55. ^Pritchard, M., "NVIDIA Launches Earth-2 Family of Open Models, the World's First Fully Open, Accelerated Set of Models and Tools for AI Weather," NVIDIA Blog, January 26, 2026. blogs.nvidia.com/...nvidia-earth-2-open-models
56. ^Robinson, C. et al., "Turkey Earthquake Report," Microsoft Research technical report MSR-TR-2023-7, February 2023. microsoft.com/...turkey-earthquake-report
57. ^NASA, "How NASA Is Testing AI to Make Earth-Observing Satellites Smarter," July 24, 2025. nasa.gov/...ake-earth-observing-satellites-smarter
58. ^Abcouwer, N. et al., "Machine Learning Based Path Planning for Improved Rover Navigation," arXiv:2011.06022, November 2020. arxiv.org/...2011.06022
59. ^Powell, M. W. et al., "MSLICE Science Activity Planner for the Mars Science Laboratory Mission," NASA Tech Briefs, September 1, 2009. ntrs.nasa.gov/...20090032136
60. ^Gaines, D. and Parjan, S., "'In OBP We Trust': Verification and Validation of the M2020 On Board Planner Flight Software," IEEE Aerospace Conference, 2024. ai.jpl.nasa.gov/...parjan-ieee2024-m2020.pdf
61. ^JPL Artificial Intelligence Group, "M2020 Onboard Scheduler," NASA Jet Propulsion Laboratory. ai.jpl.nasa.gov/...m2020-scheduler
62. ^1 ^2NASA Jet Propulsion Laboratory, "Autonomous Systems Help NASA's Perseverance Do More Science on Mars," September 21, 2023. jpl.nasa.gov/...rseverance-do-more-science-on-mars
63. ^NASA Jet Propulsion Laboratory, "NASA's Perseverance Rover Completes First AI-Planned Drive on Mars," January 30, 2026. jpl.nasa.gov/...tes-first-ai-planned-drive-on-mars
64. ^NASA Jet Propulsion Laboratory, "NASA's Perseverance Now Autonomously Pinpoints Its Location on Mars," February 18, 2026. jpl.nasa.gov/...sly-pinpoints-its-location-on-mars
65. ^1 ^2ESA, "New satellite demonstrates the power of AI for Earth observation," August 17, 2024. esa.int/...s_the_power_of_AI_for_Earth_observation
66. ^Byrne, C., "ESA's Φsat-2 Mission, Powered by Ubotica's CogniSAT, Showcases the Power of AI for Earth Observation," Ubotica Technologies, August 19, 2024. ubotica.com/...6sat-2-powered-by-uboticas-cognisat
67. ^ESA, "Φsat-2 begins science phase for AI Earth images," July 15, 2025. esa.int/...egins_science_phase_for_AI_Earth_images
68. ^1 ^2 ^3JPL Artificial Intelligence Group, "Autonomous Sciencecraft Experiment," NASA Jet Propulsion Laboratory. ai.jpl.nasa.gov/...ase
69. ^1 ^2Chien, S. et al., "Using Autonomy Flight Software to Improve Science Return on Earth Observing One," Journal of Aerospace Computing, Information, and Communication 2(4), April 2005. doi.org/...1.12923
70. ^1 ^2 ^3 ^4Alexandrov, O. et al., "Astrobee: Free-Flying Robots for the International Space Station," IEEE Transactions on Field Robotics, March 2026. ntrs.nasa.gov/...obee_IEEE_TFR_2026_compressed.pdf
71. ^1 ^2Malik, T., "It's Alive! Humanoid Robot Powers Up on Space Station," Space.com, August 22, 2011. space.com/...-station-robot-robonaut-2-power-tests
72. ^1 ^2Pearlman, R. Z., "Robonaut Returns to Earth for Repairs After Seven Years on Space Station," Space.com, May 8, 2018. space.com/...onaut-returns-earth-after-seven-years
73. ^1 ^2Badger, J., Diftler, M., Hart, S. and Joyce, C., "Advancing Robotic Control for Space Exploration Using Robonaut 2," preprint, ISS Research and Development Conference, June 2012. ntrs.nasa.gov/...20120002793.pdf
74. ^1 ^2JAXA, "First disclosure of images taken by the Kibo's internal drone 'Int-Ball'," July 14, 2017. iss.jaxa.jp/...170714_int_ball_en
75. ^1 ^2DLR, "CIMON, a 'cyber colleague' for Alexander Gerst," June 29, 2018. dlr.de/...yber-colleague-for-alexander-gerst_28678
76. ^1 ^2DLR, "CIMON-2 is on its way to the ISS," December 5, 2019. dlr.de/...20191205_cimon2-on-its-way-to-the-iss
77. ^1 ^2 ^3NASA, "Astrobee," NASA Ames Research Center. nasa.gov/...astrobee
78. ^1 ^2Clark, S., "Soyuz spacecraft, humanoid robot return to Earth after 16-day test flight," Spaceflight Now, September 6, 2019. spaceflightnow.com/...rth-after-16-day-test-flight
79. ^1 ^2 ^3Erwin, S., "Booz Allen deploys advanced language model in space," SpaceNews, August 1, 2024. spacenews.com/...-advanced-language-model-in-space
80. ^GITAI, "GITAI Passes All NASA Safety Reviews for ISS Onboard Demonstration, Ready for Launch," June 25, 2021. gitai.tech
81. ^Alamalhodaei, A., "Gitai successfully demos autonomous robot inside the International Space Station," TechCrunch, October 27, 2021. techcrunch.com/...-the-international-space-station
82. ^1 ^2JAXA Human Spaceflight Technology Directorate, "Int-Ball2 and Astrobee make their first joint appearance! The Final Run of the 5th Kibo Robot Programming Challenge was conducted on the ISS!," November 7, 2024. humans-in-space.jaxa.jp/...004234
83. ^1 ^2JAXA Human Spaceflight Technology Directorate, "Int-Ball2 ROS/Gazebo Simulator Released," March 27, 2025. humans-in-space.jaxa.jp/...004595
84. ^NASA, "NASA to Launch Human-Like Robot to Join Space Station Crew," April 14, 2010. nasa.gov/...-like-robot-to-join-space-station-crew
85. ^Rogers, J. M., "R5," NASA Technology, September 23, 2015. nasa.gov/...r5
86. ^Airbus, "Crew assistant CIMON successfully completes first tasks in space," November 16, 2018. airbus.com/...s-Release-CIMON-20181116-EN-long.pdf
87. ^Airbus, "CIMON is back on Earth after 14 months on the ISS," August 28, 2019. airbus.com/...Earth-after-14-months-on-the-ISS.pdf
88. ^DLR, "CIMON-2 makes its debut on the ISS," April 15, 2020. dlr.de/...00415_cimon-2-makes-its-debut-on-the-iss
89. ^ESA, "CIMON and Marcus Wandt," February 8, 2024. esa.int/...CIMON_and_Marcus_Wandt
90. ^Garcia, M. A., "Station Crew Demonstrates AI Assistant and Ultra-High Res Camera," NASA Space Station blog, September 19, 2024. nasa.gov/...ai-assistant-and-ultra-high-res-camera
91. ^DLR, "Roboter spielen 'Ich sehe was, was du nicht siehst' an Bord der Internationalen Raumstation" (in German), July 31, 2025. dlr.de/...-an-bord-der-internationalen-raumstation
92. ^NASA, "Look, No Hands! NASA's First Astrobee Robot 'Bumble' Starts Flying in Space," June 20, 2019. nasa.gov/...robee-robot-bumble-starts-flying-space
93. ^NASA, "Astrobee Space-Bots Mark a New Milestone in Human-Robot Teamwork," July 22, 2022. nasa.gov/...ark-new-milestone-human-robot-teamwork
94. ^1 ^2Winget, E., "Reinforcement Learning is Making a Buzz in Space," U.S. Naval Research Laboratory (via DVIDS), September 9, 2025. dvidshub.net/...547640
95. ^Mora Vargas, A., "Astrobee Contribution to Collaborative ISS Free-Flyer Robot Paper," NASA Ames Research Center, conference paper for iSpaRo 2025, June 2025. ntrs.nasa.gov/...draft_astrobee_isparo_2025.pdf
96. ^1 ^2Stanford Report, "AI advances robot navigation on the International Space Station," December 3, 2025. news.stanford.edu/...e-station-autonomous-missions
97. ^Arkisys, "Arkisys announces the return of NASA's Astrobee," EE Journal, January 21, 2026. eejournal.com/...nces-the-return-of-nasas-astrobee
98. ^Keane, T., "This 'hands-on' AI-based test project will help ensure astronaut gloves are safe in space," Official Microsoft Blog, April 4, 2022. blogs.microsoft.com/...ut-gloves-are-safe-in-space
99. ^Barr, J., "How We Sent an AWS Snowcone into Orbit," AWS News Blog, July 21, 2022. aws.amazon.com/...-sent-an-aws-snowcone-into-orbit
100. ^Booz Allen Hamilton, "Booz Allen Deploys the Power of Generative AI in Space," press release (reprinted by Silicon UK), August 1, 2024. silicon.co.uk/...e-power-of-generative-ai-in-space
101. ^Meta, "Space Llama: Meta's Open Source AI Model Is Heading Into Orbit," April 25, 2025. about.fb.com/...source-ai-model-heading-into-orbit
102. ^Booz Allen Hamilton, "Booz Allen and Meta Successfully Demonstrate AI Vision Language Model for Space," press release (reprinted by Silicon UK), April 25, 2025. silicon.co.uk/...i-vision-language-model-for-space
103. ^NASA Science, "Kepler / K2." science.nasa.gov/...kepler
104. ^NASA Science, "TESS (Transiting Exoplanet Survey Satellite)." science.nasa.gov/...tess
105. ^1 ^2Shallue, C. J. and Vanderburg, A., "Identifying Exoplanets with Deep Learning: A Five-planet Resonant Chain around Kepler-80 and an Eighth Planet around Kepler-90," The Astronomical Journal 155(2), 94, February 2018. doi.org/...aa9e09
106. ^1 ^2Hoover, R., "New Deep Learning Method Adds 301 Planets to Kepler's Total Count," NASA, November 23, 2021. nasa.gov/...dds-301-planets-to-keplers-total-count
107. ^1 ^2 ^3 ^4Valizadegan, H. et al., "ExoMiner: A Highly Accurate and Explainable Deep Learning Classifier That Validates 301 New Exoplanets," The Astrophysical Journal 926(2), 120, February 2022. doi.org/...ac4399
108. ^Valizadegan, H., Martinho, M. J. S. and Jenkins, J. M., "Multiplicity Boost of Transit Signal Classifiers: Validation of 69 New Exoplanets using the Multiplicity Boost of ExoMiner," The Astronomical Journal 166(1), 28, June 2023. doi.org/...acd344
109. ^Valizadegan, H. et al., "ExoMiner++: Enhanced Transit Classification and a New Vetting Catalog for 2-Minute TESS Data," arXiv:2502.09790, February 13, 2025. arxiv.org/...2502.09790
110. ^Hausen, R. and Robertson, B. E., "Morpheus: A Deep Learning Framework for the Pixel-level Analysis of Astronomical Image Data," The Astrophysical Journal Supplement Series 248(1), 20, May 2020. doi.org/...ab8868
111. ^Robertson, B. E. et al., "Morpheus Reveals Distant Disk Galaxy Morphologies with JWST: The First AI/ML Analysis of JWST Images," The Astrophysical Journal Letters 942, L42, 2023. arxiv.org/...2208.11456
112. ^Caulfield, B., "Stunning Insights from James Webb Space Telescope Are Coming, Thanks to GPU-Powered Deep Learning," NVIDIA Blog, June 8, 2022. blogs.nvidia.com/...ing-james-webb-space-telescope
113. ^Caulfield, B., "Making Sense of the Early Universe," NVIDIA Blog, April 23, 2026. blogs.nvidia.com/...ai-gpu-early-universe-astronomy
114. ^University of Washington Department of Astronomy, "Astronomy Highlights," May 19, 2025. astro.washington.edu/...astronomy-highlights
115. ^Rubin Observatory, "Ever-changing Universe Revealed in First Imagery From NSF-DOE Vera C. Rubin Observatory," June 23, 2025. rubinobservatory.org/...first-imagery-rubin
116. ^Harmon, S., "HEP Case Study: The Rubin Observatory and the Images Sent Around the World," ESnet, January 23, 2026. es.net/...ory-and-the-images-sent-around-the-world
117. ^Groff, A. and Suter, G., "NSF-DOE Vera C. Rubin Observatory launches real-time discovery machine for monitoring the night sky," SLAC National Accelerator Laboratory, February 25, 2026. www6.slac.stanford.edu/...overy-machine-monitoring
118. ^Rubin Observatory, "LSST Science Pipelines." pipelines.lsst.io
119. ^LSST Dark Energy Science Collaboration, "Opportunities in AI/ML for the Rubin LSST Dark Energy Science Collaboration," arXiv:2601.14235, January 20, 2026. arxiv.org/...2601.14235
120. ^Euclid Collaboration (Mellier, Y. et al.), "Euclid. I. Overview of the Euclid mission," Astronomy & Astrophysics 697, A1, 2025. doi.org/...202450810
121. ^ESA, "Euclid opens data treasure trove, offers glimpse of deep fields," March 19, 2025. esa.int/...ure_trove_offers_glimpse_of_deep_fields
122. ^ESA, "DR1 Timeline," Euclid Consortium / ESA Cosmos, updated June 15, 2026. cosmos.esa.int/...dr1-timeline
123. ^George, D. and Huerta, E. A., "Deep Learning for real-time gravitational wave detection and parameter estimation: Results with Advanced LIGO data," Physics Letters B 778, 64-70, March 2018. doi.org/...j.physletb.2017.12.053
124. ^Zevin, M. et al., "Gravity Spy: integrating advanced LIGO detector characterization, machine learning, and citizen science," Classical and Quantum Gravity 34(6), 064003, 2017. doi.org/...aa5cea
125. ^Connor, L. and van Leeuwen, J., "Applying Deep Learning to Fast Radio Burst Classification," The Astronomical Journal 156(6), 256, December 2018. doi.org/...aae649
126. ^SETI Institute, "COSMIC at the Very Large Array." seti.org/cosmic
127. ^Tremblay, C. D. et al., "COSMIC's Large-Scale Search for Technosignatures during the VLA Sky Survey: Survey Description and First Results," arXiv:2501.17997, January 29, 2025. arxiv.org/...2501.17997
128. ^1 ^2Ma, P. X. et al., "A deep-learning search for technosignatures from 820 nearby stars," Nature Astronomy 7, 492-502, January 30, 2023. doi.org/...s41550-022-01872-z
129. ^1 ^2Camporeale, E. and Berger, T. E., "Verification of the NOAA Space Weather Prediction Center Solar Flare Forecast (1998-2024)," Space Weather 23(10), e2025SW004546, October 2, 2025. doi.org/...2025SW004546
130. ^1 ^2 ^3FDL.AI, "About." fdl.ai/about
131. ^NASA, "NASA-enabled AI Predictions May Give Time to Prepare for Solar Storms," March 30, 2023. nasa.gov/...-give-time-to-prepare-for-solar-storms
132. ^Upendran, V. et al., "Global Geomagnetic Perturbation Forecasting Using Deep Learning," Space Weather 20(6), e2022SW003045, May 31, 2022. doi.org/...2022SW003045
133. ^1 ^2FDL.AI, "FDL 2024." fdl.ai/fdl2024
134. ^Jarolim, R., Veronig, A. M., Pötzi, W. and Podladchikova, T., "A deep learning framework for instrument-to-instrument translation of solar observation data," Nature Communications 16, 3157, April 2, 2025. doi.org/...s41467-025-58391-4
135. ^Jarolim, R. et al., "SuNeRF: 3D Reconstruction of the Solar EUV Corona Using Neural Radiance Fields," The Astrophysical Journal Letters 961(2), L31, January 2024. doi.org/...ad12d2
136. ^Legnaro, E., Guastavino, S., Piana, M. and Massone, A. M., "Deep Learning for Active Region Classification: A Systematic Study from Convolutional Neural Networks to Vision Transformers," arXiv:2410.17816, October 2024. arxiv.org/...2410.17816
137. ^Guastavino, S., Legnaro, E., Massone, A. M. and Piana, M., "Artificial Intelligence Could Have Predicted All Space Weather Events Associated with the May 2024 Superstorm," arXiv:2501.14684, January 24, 2025. arxiv.org/...2501.14684
138. ^1 ^2She, H., Phelps, C., Marshall, O. and Dale, E., "Multi-Layered Machine Learning for Rapid LEO Object Characterization leveraging Global Radar Data," Advanced Maui Optical and Space Surveillance Technologies (AMOS) Conference, 2024. amostech.com/...She.pdf
139. ^LeoLabs, "Radars." leolabs.space/radars
140. ^LeoLabs, "AI." leolabs.space/ai
141. ^Slingshot Aerospace, "Slingshot Aerospace and DARPA Develop New AI System Capable of Detecting Anomalous Satellites," June 5, 2024. slingshot.space/...slingshot-darpa-agatha-ai
142. ^NOAA Office of Space Commerce, "NOAA Awards TraCSS Presentation Layer Contract," November 2024. space.commerce.gov/...-presentation-layer-contract
143. ^Kurkowski, S., "Apple co-founder Steve Wozniak's Privateer comes out of stealth mode with satellite tracker 'Wayfinder'," Space Explored, March 1, 2022. spaceexplored.com/...h-satellite-tracker-wayfinder
144. ^Reuters, "Exclusive: Wozniak's space firm, Privateer, buys Orbital Insight, raises $56.5 million," May 6, 2024 (excerpt republished by Privateer). privateer.com/...ve-privateer-buys-orbital-insight
145. ^European Union, "Regulation (EU) 2021/696 of the European Parliament and of the Council of 28 April 2021 establishing the Union Space Programme and the European Union Agency for the Space Programme and repealing Regulations (EU) No 912/2010, (EU) No 1285/2013 and (EU) No 377/2014 and Decision No 541/2014/EU," EUR-Lex. eur-lex.europa.eu/...oj
146. ^EU SST, "Four EU Member States officially join the EU SST Partnership," April 2026. eusst.eu/...tes-officially-join-eu-sst-partnership
147. ^EUSPA, "EU SST at 10: Keeping watch over a crowded sky," July 9, 2026. euspa.europa.eu/...-keeping-watch-over-crowded-sky
148. ^ESA, "CREAM: avoiding collisions in space through automation," August 12, 2025. esa.int/..._collisions_in_space_through_automation
149. ^1 ^2 ^3Pultarova, T., "SpaceX Starlink satellites made 50,000 collision-avoidance maneuvers in the past 6 months. What does that mean for space safety?," Space.com, July 23, 2024. space.com/...sion-avoidance-maneuvers-space-safety
150. ^Clark, S., "SpaceX unveils first batch of larger upgraded Starlink satellites," Spaceflight Now, February 26, 2023. spaceflightnow.com/...upgraded-starlink-satellites
151. ^Pultarova, T., "Every SpaceX Starlink satellite has to dodge a collision almost weekly, and experts fear the worst," Space.com, July 15, 2026. space.com/...ost-weekly-and-experts-fear-the-worst
152. ^NASA and Space Exploration Technologies Corp., "Nonreimbursable Space Act Agreement between the National Aeronautics and Space Administration and Space Exploration Technologies Corp. for Flight Safety Coordination with NASA Assets," January 2021. nasa.gov/...nasa-spacex_starlink_agreement_final.pdf
153. ^AFP, "China slams U.S. after space station 'close encounters' with Elon Musk satellites," CBS News, December 28, 2021. cbsnews.com/...rlink-space-staton-close-encounters
154. ^Robinson-Smith, W., "SpaceX launches overnight Starlink flight as it unveils new 'Stargaze' space situational awareness system," Spaceflight Now, January 30, 2026. spaceflightnow.com/...situational-awareness-system
155. ^Holmes, M., "Eutelsat Completes OneWeb Deal," Via Satellite, September 28, 2023. satellitetoday.com/...eutelsat-completes-oneweb-deal
156. ^Airbus, "Eutelsat orders 229 additional OneWeb satellites from Airbus," September 10, 2026. airbus.com/...tional-oneweb-satellites-from-airbus
157. ^Ramon.Space, "Ramon.Space Awarded Eutelsat OneWeb Contract to Provide Cutting-Edge Satellite Onboard Digital Channelizers," May 14, 2025. ramon.space
158. ^Feldscher, J., "Project Kuiper Becomes Amazon Leo," Payload, November 14, 2025. payloadspace.com/project-kuiper-becomes-amazon-leo
159. ^Amazon, "Amazon Project Kuiper: Designing for safe, sustainable space operations," May 5, 2023. aboutamazon.com/...zon-project-kuiper-space-safety
160. ^Executive Order 13960, "Promoting the Use of Trustworthy Artificial Intelligence in the Federal Government," Federal Register 85(236), December 8, 2020 (signed December 3, 2020). govinfo.gov/...2020-27065.pdf
161. ^NASA, "NASA FY24 AI Use Case Inventory" (spreadsheet), December 2024. nasa.gov/...nasa-fy24-ai-inventory-csv-final.xlsx
162. ^NASA, FY25 AI Use Case Inventory (spreadsheet), January 2026. nasa.gov/...grid-6036111-data.xlsx
163. ^FDL.AI, "FDL History." fdl.ai/fdl-history
164. ^1 ^2JPL Artificial Intelligence Group, "CASPER," NASA Jet Propulsion Laboratory. ai.jpl.nasa.gov/...casper
165. ^ESA, "ESA Φ-lab." philab.esa.int
166. ^ESA, "Mission complete for ESA's OPS-SAT flying laboratory," June 4, 2024. esa.int/...ete_for_ESA_s_OPS-SAT_flying_laboratory
167. ^Space Development Agency, "About Us." sda.mil/...about-us
168. ^Thropp, C., "Space Development Agency Joins Space Force Team," ExecutiveGov, October 3, 2022. executivegov.com/...-agency-joins-space-force-team
169. ^Air & Space Forces Magazine, "Proliferated Warfighter Space Architecture" (status as of March 2026). airandspaceforces.com/...pwsa
170. ^Scientific Systems Company, "Scientific Systems Company, Inc (SSCI) Conducts Initial Demonstrations of AI-Enabled Edge-Based Processing On-Orbit for the Space Development Agency," February 8, 2022. ssci.com/...orbit-for-the-space-development-agency
171. ^DARPA, "Hallmark Transitions Key Strategies for Space Situational Awareness, Management," March 27, 2020. darpa.mil/...hallmark-transitions-space-awareness
172. ^Zhang, L. A. et al., "Artificial Intelligence and Machine Learning for Space Domain Awareness: Characterizing the Impact on Mission Effectiveness," RAND Corporation, RR-A2318-1, November 21, 2024. rand.org/...RRA2318-1
173. ^Tran, J. et al., "Artificial Intelligence and Machine Learning for Space Domain Awareness: The Development of Two Artificial Intelligence Case Studies," RAND Corporation, RR-A2318-2, September 30, 2024. rand.org/...RRA2318-2
174. ^Slingshot Aerospace, "AFWERX Selects Slingshot Aerospace to Track and Identify Nefarious In-Space Activities," April 2, 2025. slingshot.space/...y-nefarious-in-space-activities
175. ^Qualcomm, "The Snapdragon 801 Processor Is a Smooth Step Up from the Snapdragon 800 Processor," Qualcomm OnQ, February 2014. qualcomm.com/...snapdragon-801-mwc
176. ^Cretu, A. et al., "Benchmarking and Testing of Qualcomm Snapdragon System-on-Chip for JPL Space Applications and Missions," NASA, 2022. ntrs.nasa.gov/...20230006988
177. ^NASA Science, "Rover Components" (Mars 2020 Perseverance). science.nasa.gov/...rover-components
178. ^ESA, "LEON: the space chip that Europe built," January 7, 2013. esa.int/...LEON_the_space_chip_that_Europe_built
179. ^Paolo, F. S. et al., "Satellite mapping reveals extensive industrial activity at sea," Nature 625, 85-91, January 3, 2024. doi.org/...s41586-023-06825-8
180. ^Reuters, "US aerospace firm Maxar disables satellite photos for Ukraine," Arab News, March 7, 2025. arabnews.com/...world
181. ^National Space Facilities Control and Test Center (Ukraine), "Maxar restores Ukrainian users' access to GEGD satellite images," March 2025. spacecenter.gov.ua/...-suputnikovix-zobrazhen-gegd
182. ^Smith, S. et al., "A radio technosignature search towards Proxima Centauri resulting in a signal of interest," Nature Astronomy, October 25, 2021. doi.org/...s41550-021-01479-w
183. ^Sheikh, S. Z. et al., "Analysis of the Breakthrough Listen signal of interest blc1 with a technosignature verification framework," Nature Astronomy, October 25, 2021. doi.org/...s41550-021-01508-8
184. ^Martineau, K., "IBM and NASA release a new version of Prithvi," IBM Research, December 6, 2024. research.ibm.com/...prithvi2-geospatial
185. ^Peterson, A., "IBM and ESA open-source TerraMind, the best performing generative AI model for Earth observation," IBM Research, April 22, 2025. research.ibm.com/...nd-esa-earth-observation-model
186. ^Google DeepMind, "AlphaEarth Foundations helps map our planet in unprecedented detail," July 30, 2025. deepmind.google/...-planet-in-unprecedented-detail
187. ^1 ^2GITAI, "Space robotics start-up GITAI completes successful technology demonstration inside the ISS," October 28, 2021. gitai.tech/...iss-tech-demo-ja
188. ^1 ^2GITAI, "GITAI Completes Fully Successful Technology Demonstration Outside the ISS," March 19, 2024. gitai.tech/...nology-demonstration-outside-the-iss
189. ^1 ^2JAXA, "JAXA President Monthly Press Conference, November 2023," November 2023. global.jaxa.jp/...202311
190. ^Pearlman, R. Z., "NASA's Robonaut-2 reunited with its ride into space at Smithsonian," Space.com, October 25, 2024. space.com/robonaut-2-r2-humanoid-smithsonian-display