A sun lamp illuminates a satellite mockup covered in a gold thermal blanket. In the center stands a cup-shaped thruster whose iridescent surface shimmers with rainbow colors. A few meters away, out of frame, a camera slowly approaches and scans the scene – as in a real spacecraft rendezvous exercise. Precisely such scenes are being turned by European companies today into technologies that will surely manage traffic in Earth's orbits tomorrow.

The Luxembourg duo and the European testbed for autonomous navigation

Two companies from Luxembourg – LMO and ClearSpace – are developing autonomous navigation systems intended for orbital rendezvous and capture within European programs. To verify algorithms and sensors, they rely on the Guidance, Navigation and Control Rendezvous, Approach and Landing Simulator (GRALS), part of ESA's Guidance, Navigation and Control (GNC) testing facilities at the ESTEC technical center in the Netherlands. The GRALS environment connects two robotic arms on long rails: one carries the complete "chaser" with cameras and image processing computers, and the other holds the target – a satellite mockup with real materials and surface details. In such a laboratory, it is possible to safely perform hundreds of approaches and "flights at true scale," including rapid angle changes, extreme lighting contrasts, and complex trajectories that would be too expensive or too dangerous to experiment with in space.

LMO and partners, as part of the DIOSSA (Development of In-Orbit Servicing Space Situational Awareness Payloads) activity, are developing a system for visual perception and relative navigation during rendezvous with "non-cooperative" objects – from spent satellites to broken-off adapters and rocket stages. In parallel, ClearSpace in Luxembourg is building a portfolio of services for satellite life extension and debris removal, and is preparing the first commercial demonstrations in geostationary orbit. Their common denominator is vision-based navigation (VBN) systems and verifications in GRALS, which give the industry a controlled "space playground" before flight.

Why "seeing" in orbit is hard

In interstellar blackness, shiny materials and sharp contrasts create optical illusions. Satellites rotate, cast deep shadows, and distance and relative speed change perspective second by second. VBN systems must estimate the target's "pose" – its position and orientation in six degrees of freedom – from one to several images in real time, under unknown lighting conditions. By comparison, autonomous driving on Earth has road markings, signs, and hundreds of millions of reference examples; in orbit, this does not exist. That is why algorithms are learned and verified on a combination of synthetic data and physical mockups in the laboratory.

Mockups used in GRALS are selected and manufactured to resemble real platforms. Surfaces are covered with multi-layer insulation (MLI), replicas of antennas, sensors, and brackets are attached to them, and samples of solar cells are inserted on the sunny side. This achieves the optical representativeness needed for neural networks and classical algorithms to "see" during learning what they will actually see in orbit.

What a typical test in GRALS looks like

In the early phase of VBN testing, multi-purpose cameras record the target from greater distances, and computer vision uses contours and illuminated edges to determine direction and rough distance. As the "chaser" advances, the target resolution in pixels increases, so the system can also estimate relative orientation and angular velocities. Final checks cover extremely close approaches, when it is necessary to distinguish details like screws, slots, and thermal wrinkles that create unusual shadows. GRALS at those moments enables a completely darkened "space" chamber with a single solar light source and precise, repeatable robotic arm movements, which is crucial for methodical validation.

Besides cameras in the visible spectrum, other sensors are integrated – LIDARs, depth cameras, and even radar rangefinders – to obtain redundant data in poor lighting conditions or when the target is covered with hot MLI that creates saturation. The software then fuses measurements and makes decisions about thruster impulses: whether to brake, turn, circle the target, or retreat to a safe distance.

DIOSSA: from laboratory to space

DIOSSA is a multi-year activity supported by the Luxembourg program LuxIMPULSE. The goal is to create an autonomous "payload" – an SSA/VBN module – that can be installed on servicing spacecraft or as a secondary system on existing platforms. LMO with partners is developing algorithms for situational awareness, object detection and recognition, and robust pose estimation in all approach phases. This creates the possibility for the servicer, when found in the immediate vicinity of a decommissioned satellite, to make fast and safe decisions without constant support from Earth.

Luxembourg has systematically invested in space innovations over the past decade to attract companies dealing with in-orbit services, traffic management, and space environment surveillance. Through LuxIMPULSE, industrial development, prototypes, and demonstrators are funded, and implementation is coordinated by the Luxembourg Space Agency (LSA) in cooperation with ESA. It is within this framework that the Luxembourg presence of ClearSpace was built, which, alongside debris removal, has developed plans for extending the life of geostationary satellites.

ClearSpace and the new wave of services in the GEO belt

Geostationary orbit (GEO) is full of expensive, but technically healthy satellites that have run out of fuel. Instead of being prematurely retired to a "graveyard" orbit, servicers can catch them, stabilize them, and provide additional years of operation. ClearSpace has started the consolidation phase of the life extension mission in GEO from 2025, with the support of Luxembourg's LuxIMPULSE through an ESA contract. The plan is to develop the capability for autonomous docking to commercial platforms and safe joint flight (tandem), thereby extending operational life without building new satellites. Such services target the period between 2028 and 2030, when many of today's GEO satellites enter "retirement".

Such operations require the same fundamental capabilities as debris removal: precise visual navigation, mechanical coupling systems, and stack control algorithms after capture. Therefore, experiences from laboratories like GRALS – where high angular velocity passes, glare conditions, and avoidance maneuvers are practiced – are directly transferable to future GEO-servicers.

LMO: algorithms "eye-to-eye" with the target

LMO was created with a mission to enable satellites with a "sense of presence" – the ability to perceive and understand their environment in flight. As part of DIOSSA and other projects, the team developed target recognition methods under various lighting phases, including conditions dominated by specular highlights from MLI and deep shadows of occluded surfaces. During publicly presented test campaigns in GRALS, LMO validated approach strategies with mockups representing geostationary and communication platforms, with the goal of reliably recognizing the target type, dimensions, and status in flight.

A key outcome of these tests is mapping reliability limits: what target sizes in pixels guarantee robust pose estimation, how much noise in data is acceptable before the system decides to retreat, and which "fail-safe" maneuvers minimize collision risk. Such metrics eventually enter the operational rulebooks of future services – from deorbiting to inspection and upgrades.

From mockups to field data: how the view is "learned"

Training neural networks for VBN relies on a combination of synthetic and physical data. Synthetic scenes allow rapid coverage of a vast space of variations (lighting angles, textures, backgrounds), but physical models in the laboratory reveal "reality errors" – unexpected reflections, inaccuracies in textures, joint tolerances. Therefore, in later development phases, larger mockups are introduced into GRALS, used in final approaches where it is necessary to realistically see fine surface topography and precisely manage thrusts in very short intervals.

What "non-cooperative target" means and why it matters

Unlike cooperative objects (e.g., space stations with visual markers and ports), most older satellites have neither active orientation control nor standardized grappling points. Some rotate slowly by precession, others have worn, damaged, or partially deployed elements. VBN must first recognize what it is dealing with, estimate speeds and orientations, and only then choose an approach – from the "night" side for better contrast, at a tangential angle to avoid antennas, or above the polar axis for easier stabilization after capture. In case of dangerous resonances and unexpected flashes from MLI, the system must be ready for an automatic retreat and a new approach.

European context: Zero Debris and traffic management

ESA's Space Safety program has adopted the Zero Debris goal by the 2030s, which means a radical reduction in the creation of new fragments and active management of legacy objects. Missions like ClearSpace-1 – the first European demonstration of capturing a non-operational object – and life extension initiatives in GEO are part of the same ecosystem: prevention and rehabilitation. As the number of satellites multiplies, without autonomous inspection, avoidance, and service systems, risks would grow exponentially. In this sense, laboratories like GRALS ensure that algorithms and sensors mature in "real" conditions before flight.

Technologies behind the scenes: from calibration to certification

A successful VBN chain begins with camera calibration and accurate knowledge of optics: focal lengths, distortions, principal point shifts. This is followed by rigorous synchronization of sensors and timestamps to merge visual and inertial data. In GRALS, these processes are practiced with control over all environmental parameters – from light source intensity to linear sled speed. Finally, robustness must be proven: that the system retains performance despite sensor degradation, cosmic radiation, thermal dilations, and slow drift rotation of the target.

The certification path for flight also includes safety analyses: defining a "cis-corridor" around the target, minimum abort distances, automated procedures for retreat in case of visual tracking loss or image saturation. Such scenarios today undergo thousands of simulations and hundreds of hours of hardware-in-the-loop (HIL) testing precisely on platforms like GRALS.

Applications beyond service: from asteroids to formation flying

Although satellite maintenance and disposal are the main drivers, the same VBN principles drive other missions: precise close navigation in small body exploration, safe landing on the Moon or Mars, and formation flying of multiple spacecraft jointly carrying instruments. GRALS has in the past also served to test visual metrics used by ESA's planetary defense missions and formation flight technologies, which is why the facility is continuously upgraded with new modules, lighting configurations, and robotic capabilities.

Luxembourg's industrial impulse

Luxembourg was among the first EU countries to recognize the economic potential of in-orbit services. The combination of incentives through LuxIMPULSE, LSA support, and connections with research centers – such as SnT at the University of Luxembourg – created a climate where specialized teams for autonomy, perception, and system safety emerge. LMO in that ecosystem builds products that give satellites "sight" and a "sense of proximity," while ClearSpace from Luxembourg develops commercial operations that reduce costs for satellite owners and open the way toward a circular economy in space.

What follows: from validation to operations

The next steps are clear: finish algorithm validation on representative mockups and real hardware assemblies, select reference missions for close approach demonstration, and finally, certify approach procedures that will transition from the laboratory to daily practice. As the industry consolidates, standards for grappling points, visual markers, and common data protocols will appear, but until then VBN must remain a "polyglot" – capable of recognizing and safely capturing diverse targets without prior markings.

The path to sustainable space traffic passes through a combination of smart sensors, robust algorithms, and credible testbeds. From the gold wrinkles of MLI that confuse cameras to precise thruster impulses when "tapping the nose" of the target – technologies being practiced in Noordwijk today will decide tomorrow whether our orbits remain safe, functional, and open to new generations of missions and services.