The tip of a rocket carries a covering called the payload fairing, which protects the cargo during launch. Even on large rockets, the interior diameter is generally limited to about 4 to 5 meters. The primary mirrors of the giant space telescopes astronomers want, high-gain communications antennas, and the large structures needed for future space-based solar power satellites cannot be launched in their finished form. Traditionally, there have been two main approaches: assembling a structure on the ground, folding it for launch and deploying it in orbit, or, as with the International Space Station (ISS), carrying modules up in multiple launches and having astronauts connect them during spacewalks.
However, the more complex the hinges and drive mechanisms, the greater the risk of failure in deployable designs, and astronaut-based assembly has limits in terms of safety and operating cost. As structures grow from tens of meters to hundreds of meters, these constraints become even more severe. As a solution, technology in which robots assemble components in orbit and extend a structure to the required size has long been studied.
On November 5, 2026, the California Institute of Technology (Caltech) and the U.S. Defense Advanced Research Projects Agency (DARPA) announced that the in-space demonstrator ASTRA (Autonomous In-Space Tensioned Reflector Assembly) had successfully assembled a structure automatically in low Earth orbit, without astronaut assistance. The two organizations describe it as the first time a structure has been assembled autonomously in space.
The result, however, does not mean an antenna was completed and put into operation in space, nor that a space factory producing components from raw materials is up and running. In practice, using two ground-assembled truss bays as a foundation, the spacecraft took components from an onboard dispenser, assembled one new bay and connected it. The core of the achievement is the demonstration of a basic process: combining prefabricated parts in space to extend a structure.
What Was Actually Assembled in Orbit, and the Unfinished Antenna
The ASTRA project originally aimed to complete a circular mesh reflector antenna with an outer diameter of 1.4 meters in orbit. The structure consists of tubular carbon-fiber composite members (struts), aluminum connectors (joints), and a porous reflective membrane made with Kapton, a type of polyimide film. The design has a ring truss around the perimeter that holds the inner membrane under tension to maintain the specified parabolic shape.
But as of the November 2026 announcement, the only bay successfully assembled in orbit was a single, third bay added to the structure, out of the 11 bays planned. The demonstrator housed two truss bays assembled in advance on the ground; after reaching orbit, the restraints were released and they were pushed out of the vehicle. A robotic mechanism then took components from the dispenser, assembled the third bay and connected it to the existing structure. In other words, the demonstration did not involve assembling all 11 bays from scratch in space; it added one new bay to two ground-built bays.
According to Caltech, because the time available for work in orbit was limited, only the initial portion of the antenna was assembled. Caltech also indicated that it would be difficult to add the remaining bays and complete the 11-bay ring truss before the mission ends in December 2026. Because no finished reflector was obtained, no results have been reported on the antenna's transmit/receive performance, gain, or reflector surface accuracy.
What was confirmed is the mechanical feasibility of extending a truss structure in the space environment, with a robotic mechanism grasping individual components, placing them in position and joining them. This is an important enabling technology for building large structures in orbit, but it does not demonstrate the performance of a finished antenna.
データを表で見る
| 区画数 (区画数 (ベイ)) | |
|---|---|
| 地上先行製作 | 2 |
| 軌道上新規組立 | 1 |
| 未組立区画 | 8 |
| 設計目標全数 | 11 |
One of the 11 target bays was newly assembled in orbit. Although it fell short of a finished antenna, it is significant that the process of sequentially joining ground-manufactured components in space and adding a new bay to an existing truss worked in the actual orbital environment. Whether the same process can be repeated to complete a large structure will determine the technology's practicality.
The Gantry Robot's Design and the Ground Support Hidden Within "Autonomy"
ASTRA does not use a humanoid robot or a complex multi-jointed arm. According to DARPA materials, the demonstrator carries a gantry mechanism (a gantry manipulator) that moves along three mutually orthogonal linear axes. Rather than a general-purpose robot exploring unknown environments, the design is closer to an industrial automated machine meant to repeat predetermined assembly tasks accurately.
In the planned assembly sequence, the manipulator first takes carbon-fiber composite struts and aluminum joints from a motor-driven dispenser. It connects them while positioning them on an internal assembly plate to form one bay of the truss. Next, a net-like porous film is attached, and the completed bay is pushed outward by the assembly plate to make room for assembling the next bay. The concept was to repeat these steps in sequence and ultimately complete a ring frame made of 11 bays.
| Development stage | Test article and environment | Confirmed results | Not demonstrated |
|---|---|---|---|
| 2025 ground development (earlier paper) | 1.4 m diameter, 12-bay modular prototype. Manual assembly tests under Earth gravity and a single-axis automated fixture | Design features such as a 72-degree plate tilt angle avoided component interference; assembly and prevention of cable breakage confirmed qualitatively | Behavior in microgravity, thermal cycling and vacuum; statistical repeatability of assembly; dimensional tolerances |
| 2026 orbital demonstration (ASTRA) | Payload of about 1 cubic meter. Mounted on Vigoride 7; orthogonal three-axis gantry operated in low Earth orbit | After deploying two ground-built bays, a third bay was newly assembled in orbit | Completion of all 11 bays; overall reflector surface accuracy; radio/optical performance of the antenna; long-term durability |
| NOM4D long-term goal (concept) | Structures hundreds of meters to kilometer scale. Orbital manufacturing and assembly facilities | Conceptual design and small-scale enabling technology development and demonstrations under way | Manufacturing components from raw materials; deformation control of ultra-large structures; unmanned autonomous fault recovery |
There is also ingenuity in the material used for the reflective surface. Conventional deployable antennas have widely used meshes woven from metal wire, but these required a relatively large initial tension (prestress) to suppress creases and deformation. Supporting that tension requires a sturdy outer frame, which adds to the weight of the overall structure. So the Caltech team, working with Professor Igor Bargatin of the University of Pennsylvania and colleagues, devised a porous film structure with fine holes in lightweight Kapton film.
In this design, two opposing cable nets are pulled against each other through springs to maintain the shape of the reflective surface. Because it requires less prestress than a conventional metal mesh, the truss supporting it is expected to be lighter. However, this orbital demonstration did not publish quantitative data on specific weight reduction compared with conventional structures, nor on the surface accuracy or electrical performance of a finished reflector.
Another point of caution is the meaning of the word "autonomous" as used in the project. ASTRA required no direct work by astronauts, and its robotic mechanism carried out preset assembly motions. But the robot did not itself judge unexpected problems, devise responses and finish the job. Actual operations depended on continuous monitoring and remote intervention by the ground team.
Ground engineers sent command sequences to the spacecraft and checked the results through still images sent back from the satellite. When an unexpected problem occurred, they had to investigate the cause on the ground, correct the procedures and parameters, and then send instructions again. For this, they used a flight spare, a backup unit with the same configuration as the flight demonstrator, kept in a Caltech laboratory. The published materials contain no figures on the number of retries needed for individual motions, the number of command corrections, or the frequency of ground intervention.
Therefore, ASTRA's autonomy should be evaluated only to the extent that it can automatically execute predetermined assembly motions in space. Fully autonomous operation, in which the robot alone detects unexpected faults, determines their causes and recovers, has not been demonstrated.
Geometric Interference and the Difficulty of Tension Control Revealed in Ground Experiments
Behind the cautious operations that ASTRA's orbital assembly required are difficulties peculiar to a structure that integrates a truss with a tensioned membrane. These challenges are described in detail in a technical paper the Caltech team presented in January 2025 at SciTech 2025, the American Institute of Aeronautics and Astronautics (AIAA) annual conference. The paper by Jaden Suh, S. Dassanayake and S. Pellegrino (DOI: 10.2514/6.2025-0410) summarizes the results of prototyping and ground experiments carried out before the spacecraft was launched.
If only the truss framework were being assembled, the main challenge would be avoiding geometric interference between members and joining them in the correct positions. But with ASTRA, the thin, net-like reflective surface, which is under a certain tension when completed, must be built in at the same time as the truss. A structure partway through assembly is not as stable as the finished one, and tension can make members move in unintended directions or cause local deformation. For this reason, simply connecting parts in sequence cannot guarantee stable assembly.
On the ground, the research team ran manual assembly tests using a 12-bay modular prototype with an outer diameter of 1.4 meters. They found multiple instances in which the truss and cable net deformed in complex ways during assembly, causing members to interfere with one another. To form a circular structure, the assembly plates supporting each bay need an appropriate tilt angle. An initial 90-degree arrangement did not allow assembly, and the team compared conditions such as 30, 40 and 50 degrees. Analysis gave 30 degrees as the limiting tilt angle, but actual assembly ran into problems that adjusting the angle alone could not solve.
One problem confirmed was "plate jamming," in which a component becomes wedged in a gap in the support plates and stops moving. In addition, excessive tension concentrated in some of the cables holding the reflective surface, causing them to break before assembly was finished. The team also observed "outward kinking," in which an unfinished truss bends outward, and geometric locking, in which adjacent bays become distorted and stuck. All of these arise from the mutual influence of changes in truss shape and cable tension.
To address these issues, the ground-based functional testbed aimed at automated assembly redesigned the assembly plate tilt angle to 72 degrees, taking into account its positional relationship with the drive mechanism and component dispenser. The team also revised the work sequence to avoid interference between members and introduced a spring-based mechanism to absorb deformation. As a result, it was confirmed qualitatively that components could be assembled under certain conditions and that breakage of tension cables could be prevented.
However, what the 2025 paper shows is the feasibility of the assembly mechanism in ground experiments. It contains no statistical data on how many times the same process was run and how many times it succeeded, nor dimensional tolerances of the completed truss or error bars showing measurement uncertainty. In other words, while it was confirmed that a structure can be assembled with appropriate fixtures and procedures, reliability in the space environment and the stability of repeated operation were not guaranteed.
The unexpected problems that arose during this orbital demonstration cannot be equated with the plate jamming and abnormal cable tension confirmed in ground tests. Space has environmental factors different from a ground laboratory, such as vibration in microgravity, rapid temperature changes as the spacecraft moves between sunlight and shadow, and metal-to-metal contact in high vacuum. These may affect assembly motions, but Caltech and DARPA have not publicly disclosed the causes or mechanisms of the individual problems that occurred in orbit. As a result, it is currently impossible to judge which problems could have been predicted from ground tests and which were phenomena specific to the space environment.
About Two Hours of Contact a Day and Trial and Error with a Ground Flight Spare
ASTRA's space demonstrator was launched on March 30, 2026, from Vandenberg Space Force Base in California. It rode on SpaceX's Transporter-16 rideshare mission and headed to low Earth orbit as one of 10 payloads aboard the orbital transfer vehicle Vigoride 7, operated by Momentus. ASTRA's payload volume was about 1 cubic meter, within which the robotic mechanism, component dispenser and other equipment were housed. As the host spacecraft, Vigoride 7 provided ASTRA with power, a communications interface, cameras for external imaging, thermal control and other functions.
Once operations in orbit began, the ground team faced severe limits on communication time. Vigoride 7, orbiting in low Earth orbit, could communicate directly with ground stations only in about seven to eight windows of roughly 15 minutes each per day. In other words, the total time available to check the vehicle's state from the ground and send new commands was limited to about two hours a day.
In this intermittent communication environment, remote operation and troubleshooting from the ground take time. Each time the manipulator grasped and moved a component or joined a joint, engineers had to check still images sent from the spacecraft and judge whether the part had been placed correctly. If a positional deviation or abnormal motion was suspected, work was suspended, countermeasures were considered on the ground, and then new commands were sent. Unlike operating an always-connected robot in real time, the series of tasks had to proceed one communication window at a time.
Supporting these operations was the flight spare kept in Caltech's laboratory. In the laboratory of Professor Sergio Pellegrino, who leads the project, a ground test unit with the same configuration as the orbital demonstrator had been prepared. When a problem arose in space, engineers used this backup unit to reproduce the situation and adjusted the mechanism's operating procedures, motor drive conditions, the manipulator's travel paths and more. After verifying the corrected motions on the ground, they repeated the cycle of sending instructions to the spacecraft in the next communication window and checking the results.
Verification using the flight spare was a practical means of solving problems within limited communication opportunities. However, it differs from a controlled experiment in the strict scientific sense. Because gravity and atmosphere exist on the ground and the temperature environment is controlled, phenomena that occurred in orbit cannot be fully reproduced. The flight spare functioned less as a device for directly verifying the effects of the space environment than as a test unit for confirming in advance that corrected operating procedures work mechanically.
The short development period is also a feature of this project. According to Caltech, the time required to complete the design, manufacturing and ground testing of the complex payload and deliver it to Momentus was less than 12 months. DARPA, meanwhile, describes the span from the concept of the orbital demonstration to launch as about 15 months. The two cite periods with different start and end points, but both show that the project went from development to launch in a short time.
Successfully assembling one truss bay in orbit within a limited development period and communication environment is an engineering achievement of real significance. However, the published materials do not present quantitative data such as measured forces and torques applied to each joint, the dimensional accuracy of the completed bay, the success rate of assembly operations, or measurement errors. The main evidence for judging this result is images taken in orbit, ground operations records, and the official announcements by Caltech and DARPA. For this reason, the fact that assembly succeeded must be clearly distinguished from an assessment that practical reliability and accuracy have been confirmed.
The Full Picture of NOM4D and the Challenges Remaining Before "Kilometer-Scale" Space Construction
To evaluate ASTRA's result, it is necessary to understand the purpose of DARPA's NOM4D (Novel Orbital and Moon Manufacturing, Materials, and Mass-efficient Design) program, which has supported its development.
NOM4D, which began in 2022, is a program that fundamentally rethinks the conventional approach of manufacturing space structures on the ground and folding them for launch. During a rocket launch, structures are subjected to intense acoustic vibration and accelerations reaching several G. As a result, a space structure completed on the ground must be designed to withstand launch loads, with reinforcing members that are not necessarily needed for operation in space. Such reinforcement adds weight and has constrained the scale and mass efficiency of what can be launched.
NOM4D aims to transport raw materials efficiently from Earth and manufacture and assemble the components themselves in orbit or on the Moon. If metal pellets, resin, continuous fiber and similar materials are brought up as raw materials and processed into the required shapes in space, the constraints of carrying finished products from the ground can be eased. Because reinforcement to withstand launch loads can also be reduced, this could lead to the construction of lightweight, large-scale structures that take advantage of the microgravity environment.
Andrew Detor, NOM4D's program manager, has explained that the ultimate goal is space structures hundreds of meters to kilometers in scale. The ISS, by comparison, is a huge artificial structure with a maximum dimension of about 109 meters, but building it required dozens of launches, more than 10 years and many spacewalks by astronauts. NOM4D seeks to lay a technological foundation for manufacturing and assembling such large structures in space in the future.
Still, a large technical gap remains between what ASTRA has achieved and NOM4D's long-term goals.
First, ASTRA did not "manufacture" components in orbit. The struts and joints used in assembly were parts processed and manufactured on the ground in advance and stored in the spacecraft's dispenser. What the robot did was take them out, place them in position and join them. The process did not include forming composite tubes from raw materials or sintering metal powder to make parts.
Second, the scale of the structures differs greatly. The 1.4-meter-diameter antenna ASTRA aimed for is small enough to fit in the fairing of an existing small rocket, an order of magnitude smaller than the hundreds-of-meters to kilometer-scale structures NOM4D envisions. Even if the assembly mechanism works on a small prototype, there is no guarantee that the same method can maintain sufficient accuracy and stability when scaled up to tens of meters or more.
In large structures, expansion and contraction caused by temperature changes and minute vibrations can lead to deformation of the whole, and local assembly errors may accumulate. In reflectors and antennas, such deformation directly affects final performance. ASTRA's method of adding bays one after another is intended to extend a structure in stages, but deformation control and dimensional accuracy at larger scales have not yet been verified experimentally.
Third, technology that integrates component manufacturing and assembly in orbit has not been established. In NOM4D's third phase, besides ASTRA, a project led by the University of Illinois Urbana-Champaign has also been selected for a small-scale space demonstration. It plans to use the ISS's external experiment facilities to test a thermal curing process that cures and forms carbon-fiber composite tubes in the space environment.
Unlike the mechanical assembly ASTRA addressed, this experiment targets the process of processing materials in space to manufacture components. Although both belong to the same NOM4D program, they are not experiments that reproduce or verify each other's results. At this stage, the different enabling technologies of materials manufacturing and structural assembly are each being demonstrated independently. An integrated orbital construction system that manufactures components from raw materials and automatically assembles them into a large structure has not yet been achieved.
Caltech and DARPA cite as future applications giant radio and optical telescope primary mirrors, deep-space communications antennas, large solar panels for space-based solar power, orbital fuel depots and crewed habitation modules. All are structures that are difficult to launch finished from the ground, and being able to manufacture and assemble them in orbit could broaden design freedom. However, these are future application concepts; ASTRA's demonstration did not confirm construction capability or functionality for any of them.
According to Professor Pellegrino, the Caltech team plans to simplify and improve the assembly mechanism based on lessons learned from this operation. Going forward, it will be important to reduce faults in component supply and positioning and to build in the ability to recover from unexpected problems without ground intervention. The question is whether the effort can move from the stage of successfully assembling one bay to one in which all 11 bays are assembled stably and meet the required dimensional accuracy and structural strength.
The ASTRA demonstration has shown that robotic automatic assembly of components actually works as a path toward building large structures in space. At the same time, operation without ground-team support, the reliability of repeated work, and the accuracy and performance of a completed structure still need to be verified. What will matter in future development is whether the problems that occurred this time and how they were handled can be systematically analyzed, and the system improved so assembly can continue with less ground intervention. Can the achievement of assembling one bay in space be developed into technology for autonomously building structures hundreds of meters in scale? Judging that feasibility will require further orbital demonstrations and quantitative performance evaluation.