AstroForge is preparing to send a new kind of commercial spacecraft into deep space, combining autonomous onboard computing with a mission architecture designed to reduce the amount of control that traditionally has to come from Earth.
Image Courtesy : astroforge.com
The California-based space startup is developing DeepSpace-2, a roughly 200-kilogram spacecraft scheduled for launch in the fourth quarter of 2026. The mission is designed to travel millions of kilometers from Earth, rendezvous with a near-Earth metallic asteroid and operate with a high degree of onboard autonomy. AstroForge describes DeepSpace-2 as a low-cost, repeatable platform intended to make commercial activity beyond Earth orbit more routine.
The company's approach reflects a broader transformation taking place in spacecraft design. Instead of treating a spacecraft as a machine that receives detailed instructions from operators on Earth, engineers are increasingly giving spacecraft more ability to interpret sensor information, navigate independently and respond to changing conditions without waiting for a command from mission control.
That shift is especially important once a spacecraft travels far enough from Earth for communications delays and limited bandwidth to become operational constraints.
DeepSpace-2 is being designed around that reality. AstroForge says the spacecraft will be capable of autonomous navigation and long-range operations, with communications extending as far as 20 million kilometers. Its communications system uses omnidirectional antennas, allowing the spacecraft to maintain thrust in a particular direction while continuing to communicate with Earth rather than constantly having to point directly toward its ground connection.
The spacecraft's autonomy is part of a much larger ambition.
AstroForge is attempting to establish a commercial pathway to asteroid exploration and, eventually, resource extraction. The company's long-term business model centers on metallic asteroids containing platinum-group metals and other potentially valuable materials. But before a company can mine an asteroid, it has to solve the considerably more fundamental problem of reaching one, navigating around it and operating near an object whose exact characteristics may not be known in advance.
That is why autonomous spacecraft technology is becoming so important to AstroForge's strategy.
DeepSpace-2 is designed to travel beyond Earth's planetary gravity well, conduct a deep-space journey and eventually rendezvous with a near-Earth asteroid. AstroForge's current mission materials describe a high-resolution monochromatic camera that will capture imagery of the target's shape, surface characteristics, texture and reflectivity, potentially providing clues about its mineral composition.
The company has also described the spacecraft platform as intentionally modular. Future versions could carry optical and infrared instruments, radio-frequency systems, space-domain-awareness sensors, spectrometers and eventually mining hardware. That means the technology developed for DeepSpace-2 is intended to become a foundation for multiple types of deep-space missions rather than a one-off vehicle.
Artificial intelligence could become an important part of that architecture.
In conventional spacecraft operations, many decisions are planned on Earth and transmitted to a vehicle through a network of ground stations. That approach works extremely well when the spacecraft's environment is predictable and communications are reliable.
Deep-space operations make that model harder.
A spacecraft approaching a small asteroid may need to process new observations as they arrive. Its understanding of the target can change as the vehicle gets closer. Lighting conditions can alter what its cameras see. Navigation estimates can shift. The spacecraft may encounter conditions that were not fully represented in mission planning.
The farther away the spacecraft is, the less practical it becomes to depend on Earth for every adjustment.
AstroForge itself has described autonomous navigation as a fundamental requirement for its deep-space ambitions. In a July 2026 discussion about the broader implications of its technology, the company said a spacecraft operating far beyond established navigation infrastructure needs to estimate its own position, understand the movement of nearby objects and respond to deviations from its expected trajectory.
That is where increasingly sophisticated onboard AI can become useful.
Rather than simply following a predetermined sequence of commands, an autonomous spacecraft can potentially process sensor information and make decisions locally within predefined constraints.
The distinction is important. An autonomous spacecraft does not necessarily mean an AI system has unlimited authority to control every aspect of the mission. Spacecraft designers can establish safety boundaries, predefined operating modes and conventional control systems while using AI for particular perception, navigation or decision-making tasks.
That layered approach is likely to remain important as AI becomes more deeply integrated into space systems.
The idea of putting neural networks directly into spacecraft control systems is already moving beyond laboratory research. In July 2026, the Air Force Research Laboratory announced that it had demonstrated autonomous control of a satellite bus using a neural network. According to AFRL, the system was able to control the spacecraft's orientation in orbit without human intervention during the demonstration.
NASA is also developing technologies aimed at making spacecraft more self-reliant. Its AstroNav system, being developed for a future CAPSTONE 02 mission, is intended to enable real-time autonomous navigation by allowing spacecraft to determine their position and control their trajectory without relying solely on Earth-based navigation.
The broader trend suggests that spacecraft autonomy is moving toward a more intelligent model.
Instead of merely automating a few repetitive commands, future systems could combine navigation algorithms, machine-learning models, sensor fusion and conventional flight software to give spacecraft greater ability to react to their environment.
For a company such as AstroForge, that capability could be particularly valuable because its business model depends on making deep-space missions cheaper and more repeatable.
Historically, interplanetary spacecraft have required enormous amounts of mission planning and ground support. Every additional spacecraft capability can require additional personnel, communications infrastructure and operational complexity.
AstroForge is attempting to change that economics.
The company's June 2026 description of DeepSpace-2 emphasized a philosophy of low-cost, rapidly iterated spacecraft that can fly repeatedly rather than treating every deep-space mission as a unique national-scale project. The company says future versions of its platform are intended to support missions ranging from resource prospecting and scientific exploration to planetary defense and other commercial payloads.
That philosophy could make autonomy more than a technical feature.
It could become part of the business model.
If spacecraft can perform more functions without constant human intervention, a company could potentially operate missions with smaller ground teams and reduce the amount of infrastructure required for every individual vehicle. More automation could also allow a spacecraft to respond to conditions faster than an Earth-based team could, particularly when communication delays become significant.
But there is an important distinction between autonomous operation and fully independent decision-making.
DeepSpace-2 is not being presented as a spacecraft that simply receives an objective from Earth and then decides everything else for itself. AstroForge's public materials describe a vehicle with autonomous navigation, onboard computing, precision guidance and control, and deep-space communications. The spacecraft is still part of a mission architecture involving Earth-based operations.
That distinction matters because spacecraft are unforgiving environments.
A software error that might be recoverable on Earth can become catastrophic hundreds of thousands or millions of kilometers away. A bad navigation decision can place a spacecraft on an unrecoverable trajectory. An incorrect interpretation of sensor data can cause a vehicle to miss its target or consume valuable propellant.
For that reason, AI-based spacecraft systems will likely need multiple layers of validation and redundancy.
AstroForge has already experienced the realities of operating deep-space hardware. Its first deep-space mission, Odin, launched in February 2025 with the objective of capturing imagery of asteroid 2022 OB5. The company reported that it successfully traveled beyond the Moon and returned some signals, while also identifying communications, solar-power and deployment issues that informed the design of DeepSpace-2.
The company has explicitly described DeepSpace-2 as a product of those lessons.
Its spacecraft is substantially larger than Odin, with a mass of roughly 200 kilograms, substantially greater available power and electric propulsion using Hall-effect thrusters. The current vehicle is designed around approximately 2 kilowatts of solar power and up to 5 kilometers per second of delta-v.
One of the most interesting components is its onboard computing hardware.
AstroForge lists an NVIDIA Jetson AGX Xavier Industrial system among DeepSpace-2's avionics. Jetson is an edge-computing platform designed for running AI and other computationally intensive workloads locally, which makes it well suited to a spacecraft architecture where certain calculations need to happen onboard rather than being continuously transmitted to Earth.
This reflects an increasingly important concept in space technology: edge AI.
Instead of sending raw sensor information to Earth and waiting for a ground computer to process it, a spacecraft can analyze some of that information locally.
A camera could potentially identify relevant features. A navigation system could process observations and update a spacecraft's estimated trajectory. A vehicle could determine whether sensor readings are consistent with expected conditions before deciding whether to continue, change operating mode or request assistance from Earth.
The advantage is speed and reduced communications demand.
The tradeoff is that onboard systems have limited computing resources, power and opportunities for software updates. Spacecraft designers therefore have to be extremely careful about what computational workloads they place on a vehicle.
That makes the development of efficient AI models particularly important.
The concept of a transformer-based model controlling or assisting a spacecraft is especially intriguing because transformers have become the dominant architecture behind many modern AI systems. They are capable of processing sequences of information and identifying relationships within those sequences, which has made them successful in language, vision and multimodal AI.
Applying transformer-style models to spacecraft operations, however, is a fundamentally different challenge from using one to write text.
A spacecraft requires predictable behavior, bounded failure modes and extremely high reliability. Engineers cannot simply assume that a model's most likely output is always the correct action.
The practical future of AI in spacecraft may therefore involve specialized models operating alongside traditional guidance, navigation and control software rather than replacing those systems entirely.
The AI might interpret sensor information while a conventional control system determines whether the proposed action falls within safe limits.
That kind of architecture could allow spacecraft to benefit from machine learning without handing unrestricted control to a general-purpose model.
AstroForge's broader mission architecture is already designed around autonomous operation. DeepSpace-2 is intended to spend up to two years operating on a mission that will take it as far as 20 million kilometers from Earth. The spacecraft is expected to reach the vicinity of the Moon shortly after launch, perform a lunar flyby and then enter deep-space cruise before eventually reaching its asteroid target.
That distance makes autonomy especially meaningful.
Even at the speed of light, communications over millions of kilometers introduce delays. And communication availability can be affected by spacecraft orientation, antenna geometry, ground-station access and other operational constraints.
An intelligent spacecraft cannot afford to behave like a remote-controlled drone waiting for every instruction.
It needs to keep itself alive.
It needs to understand where it is.
And increasingly, it needs to understand what is happening around it.
That is the direction in which AstroForge's spacecraft architecture is moving.
The company's ultimate objective is even more ambitious. AstroForge wants to build a commercial system capable of repeatedly reaching asteroids and eventually extracting resources from them. Its June 2026 roadmap described a future in which lower-cost spacecraft could enable more frequent missions and expand access to deep-space exploration beyond government agencies.
Autonomy could be one of the technologies that makes that economic model possible.
If every deep-space spacecraft requires a massive team of human operators making constant decisions, the cost of operating a fleet could remain extremely high. If spacecraft can increasingly handle routine navigation, observation and other operational decisions themselves, companies could potentially operate more vehicles without scaling ground operations at the same rate.
That is a particularly important consideration if asteroid mining ever progresses from individual demonstrations to a fleet-based industry.
Mining multiple asteroids would require spacecraft to perform repetitive operations across enormous distances. The economics would look very different if each mission required the same level of human attention as a flagship planetary mission.
Autonomous systems could provide the technological foundation for scaling.
But DeepSpace-2 still has to demonstrate that the concept works in the real environment.
AstroForge has not yet completed the mission. The spacecraft is currently being prepared for launch in the fourth quarter of 2026, according to the company's latest public mission information.
That means some of the most important questions remain unanswered.
How effectively will the autonomous navigation system perform once the spacecraft leaves Earth's familiar orbital environment? How well will its sensors characterize a small asteroid? How much of the mission can genuinely be conducted without human intervention? And how will its onboard computing systems behave over an extended deep-space journey?
Those questions will ultimately matter more than the marketing language surrounding autonomous spacecraft.
If DeepSpace-2 successfully demonstrates reliable autonomous operation at deep-space distances, however, it could provide a useful blueprint for a different class of commercial spacecraft.
The significance would extend beyond asteroid mining.
The same technologies could potentially support scientific probes, planetary-defense missions, space-domain-awareness systems and commercial payload platforms that need to operate far from Earth with limited communications.
AstroForge itself has highlighted those potential applications. The company argues that the capabilities being developed for asteroid mining—autonomous navigation, rendezvous, proximity operations and operation around poorly characterized objects—could have uses in science, planetary defense and national security.
The larger shift is therefore not simply about putting AI onto a spacecraft.
It is about changing where decisions are made.
For much of the space age, Earth has remained the center of spacecraft decision-making. Engineers on the ground have interpreted telemetry, planned trajectories, analyzed images and transmitted instructions.
The emerging generation of autonomous spacecraft is beginning to reverse that relationship.
Earth still establishes the mission.
But the spacecraft increasingly has to figure out how to execute it.
AstroForge's DeepSpace-2 is preparing to test that philosophy on one of the most challenging environments imaginable: a journey millions of kilometers from Earth toward a small, distant asteroid.
If the spacecraft can navigate, communicate, observe and operate reliably with increasingly intelligent systems onboard, it could demonstrate that the next era of commercial space exploration may not be defined solely by cheaper rockets and smaller spacecraft.
It may also be defined by spacecraft that can increasingly operate on their own.