In orbit, the Astra robot did not complete the 1.4-metre reflector antenna planned for the experiment. It did, however, build its third section, after receiving two sections already assembled on Earth. This limited demonstration, conducted by Caltech as part of the US NOM4D programme, establishes a first concrete step towards the automated assembly of space structures, not the ability to build giant installations today.
The aim is to manufacture objects in orbit that are too large to be launched in one piece. But between a section assembled in an experiment and an antenna several hundred metres across, there are still design, reliability and operational problems to solve.
An antenna built in stages
The Astra device, designed by Caltech with funding from the US agency DARPA, was launched in late March 2026 on SpaceX’s Transporter-16 mission. It travelled aboard the Vigoride 7 orbital vehicle, operated by Momentus. The device, about one cubic metre in size, contained a robotic manipulator, parts dispensers and two pre-assembled sections of the structure.
Once in low Earth orbit, the fastenings used to protect the equipment during launch were released. The two sections that had already been built were pushed out of the device, and the robot then assembled a third section, called a ‘bay’ in Caltech’s documentation. A bay is a repeating segment of the supporting structure, made up of tubes and joints.
The planned antenna was to consist of eleven bays connected to a reflective surface. At the time of the report, the assembly remains incomplete. Caltech considers it unlikely that the full assembly will be completed before the end of the mission, scheduled for December 2026. The distinction matters: the experiment demonstrated the autonomous assembly of a section in orbit, but not the complete construction of the 1.4-metre antenna.
The assembled components include composite tubes, aluminium joints and a thin Kapton reflective surface. The robot moves the parts, positions them and connects them on an assembly plate. The reflective surface is designed as a perforated film stretched over the structure. According to Caltech, this solution can reduce the required pre-tension compared with a metal mesh, opening the way to lighter structures.
Why assemble rather than unfold?
Large space structures are currently constrained by the dimensions of a rocket’s fairing, the casing that protects the payload during launch. Once in orbit, they can be deployed if they have been folded, or assembled from large modules sent up over several launches. In both cases, design choices are linked to what can be transported and handled.
Piece-by-piece assembly in orbit offers another option: send the components separately, then build on site a structure that could not be launched already formed. This method could be of interest for large radio antennas, reflectors for telescopes or extensive solar panels. These are envisaged applications, not equipment already built by Astra.
The NOM4D programme, launched by DARPA in 2022, is exploring methods of manufacturing and assembly in space. Its long-term ambition is to create structures several hundred metres, or even several kilometres, in size. The programme’s lead, Andrew Detor, nevertheless points out that no structure of this size can currently be built using this approach. Astra is testing one technical building block of this ambition, on a much more modest scale.
Autonomy does not rule out work on the ground
The robot carried out the assembly operations without astronaut intervention, but the operation was not independent of all human supervision. Ground teams sent commands and examined the images they received to monitor progress. Caltech says communications were possible during seven or eight daily windows of about fifteen minutes each.
This pace requires a particular way of working. When a problem arose in orbit, engineers could try to reproduce it using a backup version of the device kept in the laboratory. They would then test a solution on Earth before sending new instructions. Here, autonomy means that the robot carries out a sequence of manipulations in an environment where communications are intermittent, not that it decides on its own objective or repairs itself without help.
This stage in orbit also served a different purpose from the laboratory tests. The project’s first two phases focused on design and ground-based testing; the space demonstration was intended to check whether the system could operate in its operational environment. The teams encountered difficulties and had to adapt their procedures. The value of a test of this kind lies precisely in the gap between a sequence planned on the ground and its actual execution in orbit.
A proof of concept, not a space factory
Caltech describes the result as proof that the principle of assembly is feasible. DARPA sees it as a possible way to design future large, lightweight structures. Both formulations describe technical potential; they do not mean that a giant antenna, a habitat or a space-based solar power station can be built using the current device.
To reach this scale, many more sections would have to be assembled, the mechanical integrity of large structures guaranteed and the supply of parts organised. The operations would also have to be controlled and the reflective elements integrated. Astra provides practical experience in repeatedly handling components and assembling a section in orbit. What comes next will depend on more advanced systems and further demonstrations.
Success is therefore measured less by the size of the antenna than by the operation now tested in space: a robot picked up components, connected them to build a section and deployed part of the reflective system, with no astronaut on board to carry out the assembly. It is a milestone towards structures designed to be assembled beyond the limits of launch, not yet their standard method of construction.
References
Dextre and Canadarm2 on the International Space Station, in July 2025. Illustrative photo of orbital robotics.
Credit: NASA Johnson Space Center — Wikimedia Commons — Public domain.
