On March 30, 2026, a few hours before sunrise at Vandenberg Space Force Base, a Falcon 9 lifted a box of loose parts into low Earth orbit. The box was about one cubic metre. Inside sat composite fibre tubes, aluminium joints, two finished truss bays, and a small gantry robot that had never worked in vacuum.
Six months later, that box had built a third bay of a 1.4-metre reflector antenna. No astronaut touched the pieces. The robot reached into motorised dispensers, placed struts on an assembly plate, locked them into custom nodes, and pushed the finished section out into the dark. Ground controllers watched through 11 onboard cameras and sent commands through a host spacecraft.
Three bays of eleven were completed. Two came up prebuilt from Earth. One was made in orbit, by a machine, from parts. That is the whole of the physical result, and it is the first time a structure has been assembled this way in space.
A box of parts, then a bay
ASTRA stands for Autonomous In-Space Tensioned Reflector Assembly. Caltech designed and built it under DARPA's NOM4D programme, which began in 2022 with the ambition to ferry raw materials into orbit and manufacture there. The payload flew as one of 10 customers on Momentus's Vigoride 7 service vehicle, itself a rideshare passenger on SpaceX's Transporter-16 mission.
The hardware is deliberately plain. DARPA calls the builder a relatively simple gantry, able to move along three linear axes. It picks up struts and joints, joins them into identical bays, and attaches a perforated Kapton film that forms the reflective surface. Springs keep the film under tension. Each bay is then expelled, and the cycle repeats.
A complete antenna needs 11 bays. The plan was to build nine of them in orbit after deploying the two prebuilt sections. Time ran out before the second orbital bay was finished.
The result deserves a place in the record. It does not yet deserve the 100-metre language that arrived with it. A repeatable sequence for one bay is a real engineering milestone. A manufacturing process for a 100-metre dish is a different claim, and the demo does not settle it.
Fifteen minutes a day
The constraint that shaped the mission was not robotics. It was contact time.
ASTRA spoke to the ground through the Momentus vehicle in brief windows. Space.com reported roughly 15 minutes a day of communication. Orbital-wire counted seven or eight passes, each near 15 minutes. Either way, the team could not watch the robot work in real time. They sent a command block, waited for the next pass, downloaded photographs, and checked the outcome against a ground twin built in Sergio Pellegrino's laboratory at Caltech.
That twin mattered. When something behaved differently in orbit, engineers reproduced the problem on the ground, tested a workaround, then uploaded it. The method is ordinary. The environment is not. There is no service call, no hand on the hardware, and no way to reach in when a strut sits a millimetre off.
Caltech says the team hit time limits and built only the initial section of the antenna. The mission ends in December. Finishing the full 11-bay truss now looks unlikely, and the team says so. Eight bays remain unbuilt.
This is where the story becomes more useful than the press release. A first demonstration proves that parts can be handled and joined in orbit. It does not prove that the sequence is reliable enough to repeat dozens of times without a human in the loop. Reliability is the product. The robot is only the instrument.
The 100-metre claim is doing more work than the demo
DARPA's programme manager, Andrew Detor, framed the target plainly. "By 'very large,' I mean hundreds of meters or kilometers in scale," he said. "Nothing approaching that scale can be achieved today."
He is right about the gap. The current demo spans 1.4 metres. Momentus describes the 1.4-metre dish as a proof of concept for a 100-metre dish. That is a 71-fold increase in diameter, and the physics does not scale linearly. Stiffness, thermal distortion, pointing accuracy, and the mass of the support structure all move in the wrong direction as the aperture grows. A truss that holds its shape at 1.4 metres can sag, twist, or ring at 100.
There is also a documentation gap. Caltech and DARPA announced the result on November 5. As of publication, neither had linked a technical report with the flight data. The claim of a first comes from the organisations that ran the experiment, which is normal for a demo and still worth stating. No independent test has shown the method at larger scale.
This is the first time that a structure has been autonomously constructed in space.— Sergio Pellegrino, professor of aerospace and civil engineering, Caltech
Autonomy itself needs a qualifier. Commands came from Earth. Engineers checked photographs between passes. Neither Caltech nor DARPA has published a breakdown of how much of the work the robot decided on its own and how much depended on ground direction. The machine executed a prepared sequence; the sequence was human. That is a meaningful step toward orbital construction. It is not a self-directed factory.
The honest read is narrower and stronger. ASTRA shows that a fixed set of parts, a modest gantry, and a repeatable joint design can produce a structure in orbit without astronauts. The next version has to be simpler, lighter, and tolerant of the errors that accumulate over many cycles. Caltech says exactly that.
What Momentus sells while the truss waits
Momentus provided the host vehicle, power, communications, and imaging. That role is easy to overlook and hard to replace. A payload builder with a good idea still needs a ride, a bus, and a data path. Vigoride is that layer.
The company says Vigoride 7 has met all mission objectives since launch. Vigoride 8 is scheduled for 2027 and is fully booked with NASA payloads. Vigoride 9 is planned for late 2027 and still has capacity. Momentus is also hosting an autonomous welding payload for ARCSPACE on that later flight, a separate bet on in-orbit repair and manufacturing.
The commercial logic is straightforward. In-space assembly is a service business before it is a product business. Customers buy integration, power, comms, and a known launch slot. The robot is the customer's problem. The platform is Momentus's. A hosted payload also spreads the cost of the bus across several customers, which matters when the first buyer is a government programme. Every successful hosted demo makes the next customer's procurement easier, even when the demo itself stops at one bay.
The next milestone is a material, not a robot
NOM4D has one more flight before it closes. A University of Illinois Urbana-Champaign team will send a payload to the International Space Station later this year to test a space-compatible resin. The goal is to polymerise a high-strength composite structure in orbit with low energy input. That is a materials problem, not an assembly problem.
The pairing matters. ASTRA answered a mechanical question: can a machine join parts into a stable frame? The Illinois experiment asks whether the material itself can be made in space. Large orbital structures need both. A truss built from Earth-made tubes still pays the launch mass penalty. A truss grown or cured in orbit changes the economics.
The economics follow the geometry. A folded reflector has to survive launch loads, so its hinges, ribs, and tensioning hardware are sized for the worst case on the way up. An assembled reflector carries those loads as a structure in orbit, where stiffness and thermal behaviour can drive the design instead. That is the theoretical saving. It becomes real only when the bay sequence is reliable, because a failed joint at 400 kilometres is not a warranty event. It is a total loss of the aperture.
For now, the scoreboard is simple. One bay built in orbit, eight left unbuilt, and a December deadline that will arrive before the truss is finished. The team calls the method viable and says the design needs simplification. Both statements can be true at once.
What comes next is not a bigger dish. It is a cheaper, more repeatable bay, then a second one built without a ground rehearsal, then a structure that no rocket could have carried. That sequence will take years. The first step is now on the record, at 1.4 metres and three bays, with the hard part still ahead.