Can the UK’s AI Roadmap for Space Turn Trusted Autonomy Into a Competitive Advantage?

As an Amazon Associate we earn from qualifying purchases.

Key Takeaways

• The UK roadmap prioritizes trusted, efficient space AI over competition in computing scale alone.
• Commercial progress depends on usable data, mission assurance, procurement, and paying customers.
• The 2036 ambitions are proposed outcomes, not evidence that funding or deployment is guaranteed.

An AI Roadmap for Space Built Around Practical Advantage

On October 27, 2026, the UK Space Agency published its Artificial Intelligence Roadmap for Space, setting out strategic priorities for 2026 through 2036. The document addresses a practical industrial question: how a country with substantial research and engineering capabilities can compete when others command greater computing resources and larger space infrastructure investments. Its proposed answer centers on dependable systems, efficient processing, and services that turn space data into useful decisions.

Developed by Space Professionals Partnership Ltd for the UK Space Agency and the Department for Science, Innovation and Technology, the roadmap draws on consultation with more than 100 stakeholders. Participants included established aerospace companies, smaller businesses, universities, funding organizations, and research bodies. Their contributions informed a framework linking technological capabilities to mission requirements and potential economic value.

The document’s organizing principle is selective competition. It argues that the UK should concentrate on artificial intelligence (AI) that can be trusted in demanding environments, demonstrated against clear requirements, and operated within tight resource limits. This includes software on satellites, analytical services on the ground, autonomous spacecraft functions, and tools for manufacturing and launch operations.

That emphasis matters because space presents a different operating environment from a conventional online service. A spacecraft may have limited electrical power, intermittent communications, restricted computing capacity, and little prospect of physical repair. A faulty recommendation can also have consequences beyond an inaccurate answer: it could affect instrument scheduling, navigation, communications, or the health of an expensive asset.

The roadmap treats AI as part of a complete mission and service architecture. Algorithms require suitable hardware, reliable data, testing facilities, skilled personnel, accountable operators, and customers with a reason to pay. Improving a model in isolation does not establish that it can deliver a dependable service.

The document should also be read according to its actual status. It is a strategic framework containing recommendations, development pathways, and suggested measures of success. Its proposed national capabilities and coordination arrangements are not, by their appearance in the roadmap, funded programs or completed procurements. Its value lies in identifying where investment and institutional changes could make a difference, together with the conditions needed to turn those proposals into operational results.

Eight Capabilities Connect AI Research to Space Missions

The roadmap selects eight technology and functionality categories from a broader framework of 12. These categories overlap, reflecting how operational systems combine methods rather than relying on one isolated technique. They cover computer vision; simulation and testing; analytics, forecasting, and discovery; machine learning; planning, optimization, and scheduling; autonomous or agentic behavior; human-AI interaction; and edge computing.

Computer vision extracts information from images. In space applications, that can mean identifying objects in Earth imagery, classifying land cover, or helping a spacecraft interpret the appearance of another object during an approach. Its value depends on the quality of the observations and the ability to distinguish meaningful features from misleading patterns.

Simulation and testing support development before hardware is committed to flight. A digital twin is a model of a physical asset or system that can incorporate relevant operational data. Such models can help engineers examine faults, compare operating choices, and test software under conditions that would be expensive or dangerous to reproduce physically.

Analytics and forecasting identify unusual behavior, estimate future conditions, and support diagnosis. Machine learning provides methods for developing capabilities from data, including approaches that seek to learn from limited examples. The roadmap’s separate treatment of these categories is a planning device, rather than a claim that analytics and machine learning are unrelated disciplines.

Planning and scheduling determine how limited resources should be allocated. A spacecraft may need to balance observation requests against power, memory, communications windows, and thermal constraints. Autonomous systems extend this process by selecting and carrying out actions within an assigned scope. Human-AI interaction addresses how operators understand those actions, intervene when necessary, and maintain effective oversight.

Edge computing places processing close to the source of data, including aboard a spacecraft. It is a deployment approach rather than a single learning method. New Space Economy’s explanation of different AI workloads provides useful context for separating model training, model execution, and the infrastructure needed to support each.

The roadmap also emphasizes hybrid physics-informed AI, which combines learning methods with knowledge of physical behavior. Its appeal is straightforward: spacecraft motion, heat flow, and power systems follow physical constraints that should inform predictions. However, incorporating physics does not automatically make a model reliable. Its assumptions, training data, and behavior outside tested conditions still require examination.

These eight categories provide a common vocabulary for investment decisions. Their practical purpose is to connect a mission problem to the methods most likely to solve it, rather than encourage adoption of AI simply because a particular technique attracts attention.

Earth Observation Must Deliver Decisions That Customers Can Use

Earth observation (EO), the collection of information about the planet through remote sensing, receives prominent treatment in the roadmap. The proposed investment theme combines observation with climate intelligence and decision support. Its economic premise is that customers often need an assessment of conditions or risks more than they need an additional image.

The distinction affects the entire service. A flood-related product must do more than identify water in a satellite scene. Its usefulness depends on observation timing, the treatment of uncertainty, compatibility with other information, and delivery through a workflow that emergency planners or infrastructure operators can use. Similar requirements apply to agriculture, environmental monitoring, and assessments of physical assets.

The roadmap identifies computer vision, predictive modeling, data assimilation, and combinations of sensor types as relevant capabilities. Optical images, radar observations, and other measurements can provide complementary information. Combining them can improve interpretation, but also introduces questions about alignment, differing resolutions, calibration, and the reliability of the resulting assessment.

New Space Economy’s discussion of satellite data analytics explains the broader processing chain between an observation and an analytical product. Its coverage of answers as a service addresses the associated business shift toward delivering information that fits a customer’s decision.

The European Space Agency’s Φsat-2 mission supplies a concrete engineering example. Launched on August 16, 2024, the satellite was designed to demonstrate applications that process imagery onboard, including cloud detection, vessel detection, and image compression. The mission illustrates how some interpretation can move closer to the sensor instead of requiring every processing step to take place after transmission to Earth.

The commercial implications require more caution than the technical possibility alone. Faster processing does not necessarily produce a faster useful answer if the observation arrives too late, communications remain constrained, or customers cannot integrate the result. Nor does a successful demonstration establish demand for a recurring subscription.

The roadmap proposes a coordinated national EO capability involving multiple providers rather than one organization supplying every service. It also identifies public-sector demand as a means of supporting commercial growth. This approach could connect satellite operators, analytical businesses, domain specialists, and users around defined requirements.

A reasonable implementation test would be whether those arrangements produce repeat purchases and dependable performance. Relevant measures include delivery time, error rates, customer retention, and the cost of producing an accepted result. Counting images processed or models developed would capture activity, but would provide an incomplete account of economic value.

Security, Navigation, and Communications Require Accountable Decisions

Three investment themes address space domain awareness, resilient positioning and timing, and secure satellite communications. Their common concern is the ability to interpret changing conditions and maintain essential services. The roadmap presents AI as a tool for improving those capabilities, with assurance and human accountability remaining important.

Space domain awareness (SDA) concerns understanding objects, activity, and conditions in space. Within that field, space surveillance and tracking supports the detection and monitoring of objects and their orbits. The roadmap proposes AI-assisted sensor fusion, anomaly detection, orbit analysis, and decision support.

These functions depend on the quality and continuity of observations. Combining data from different sensors can improve understanding, but a system must also account for conflicting measurements and incomplete coverage. Detecting an unexpected maneuver is different from establishing its purpose. The roadmap’s interest in intent inference should be understood as probabilistic analysis, not a capability to determine another operator’s intentions with certainty.

Positioning, navigation, and timing (PNT) services support the determination of location, movement, and time. The roadmap recommends AI methods for detecting interference, identifying suspicious signals, and combining information from multiple sources. Jamming interferes with reception; spoofing attempts to mislead a receiver through deceptive signals.

An AI system that recognizes a suspicious pattern can contribute to resilience, but recognition is only one step. The wider system needs an appropriate response, a reliable way to communicate uncertainty, and access to alternatives when the primary signal cannot be trusted. Better detection alone cannot guarantee uninterrupted navigation or timing.

Satellite communications introduces a related resource-management problem. Demand changes, links vary, and operators must manage network performance and security. The roadmap identifies applications such as traffic prediction, beam management, spectrum monitoring, and automation of ground operations.

The proposed benefit is more responsive control of complex networks. The associated risk is that an automated decision could distribute an error across interconnected systems. An implementation needs limits on what the software may change, records of its decisions, and defined recovery arrangements.

Across these themes, the roadmap emphasizes sovereign control over important decision functions. That does not necessarily require domestic ownership of every satellite, chip, or dataset. It does require attention to who can inspect the system, maintain it, authorize changes, and continue operating when external services become unavailable.

For international partners and commercial customers, those requirements can also support trust. A service becomes more credible when its operator can explain the evidence behind an alert, describe its limitations, and demonstrate how responsibility is assigned when conditions become uncertain.

Autonomous Operations Extend From Factories to Deep Space

The roadmap’s space infrastructure theme covers manufacturing, spacecraft operations, in-orbit services, and exploration. It connects these activities through a shared objective: reducing the amount of detailed intervention needed to keep complex systems productive. The proposed gains include more efficient operations, better fault management, and longer useful mission lifetimes.

Autonomy is best understood as a set of specific permissions and capabilities. Software that schedules an instrument does not possess the same authority as software controlling a close approach to another spacecraft. Both can be described as autonomous, but their hazards, evidence requirements, and supervision needs differ substantially.

In-orbit servicing, assembly, and manufacturing (ISAM) makes these distinctions particularly important. Inspecting another spacecraft, approaching it, handling components, and performing repairs require different combinations of sensing, planning, control, and mechanical capability. Progress in one activity does not establish readiness for all the others.

The roadmap proposes combining AI-enabled operations with digital twins and predictive maintenance. Operational measurements could help identify equipment behavior that merits attention, and simulations could support assessment of possible responses. The useful outcome is better management of a physical system, rather than a more elaborate digital representation for its own sake.

Exploration provides a clear reason for greater local decision-making. Communications with Earth can be delayed or unavailable, limiting the usefulness of continuous ground direction. NASA described an operational example in July 2024: software supporting Perseverance’s PIXL instrument could examine measurements and select areas for additional investigation without consulting mission controllers during that process. That is a bounded scientific capability, rather than evidence of a rover independently setting its entire mission.

New Space Economy’s treatment of NASA’s technology priorities places autonomy within the wider requirements of exploration infrastructure. Computing must work alongside power, communications, thermal control, mobility, and maintenance. This provides useful context for the UK roadmap’s emphasis on integrated capabilities.

Launch and spaceport operations form a separate investment theme. The roadmap identifies weather analysis, trajectory optimization, maintenance, range monitoring, and operational simulation as possible applications. These tools could support better planning and reliability, but their contribution must be measured within the economics of the complete launch service.

Software improvements cannot, by themselves, resolve hardware reliability problems, weak demand, or inadequate infrastructure. The appropriate commercial test is whether an application improves a defined operational outcome after development, testing, integration, and support costs are included. That discipline helps distinguish a useful investment from a demonstration whose benefits remain difficult to realize.

Resource-Efficient Computing Fits the Conditions of Space

The AI roadmap for space treats resource efficiency as both an engineering necessity and a possible industrial advantage. Spacecraft cannot assume the power supply, cooling arrangements, maintenance access, or hardware replacement cycle of a terrestrial computing facility. Models must be designed for the environment in which they will operate.

Edge AI processes data near its source. On a satellite, that may involve identifying an observation worth transmitting, compressing information, or detecting a condition that requires a prompt response. The benefit depends on how much useful work can be completed within the available power, memory, processing time, and communications capacity.

Training and inference should be distinguished. Training develops a model using data; inference applies that model to new inputs. A mission can train a system on Earth and deploy a smaller, controlled version onboard. It does not need to reproduce the full training infrastructure in orbit to gain value from AI.

The roadmap identifies model compression, low-power embedded learning, radiation-tolerant hardware, and alternative computing architectures as investment areas. Some are closer to practical deployment than others. Their inclusion establishes research and development priorities, not a finding that all are equally mature or suitable for immediate procurement.

Reliability also depends on the data encountered after deployment. ESA-supported research conducted from April through June 2026 examined compact models using real Φsat-2 imagery and sensor-related disturbances. The project emphasized that good performance on clean images does not automatically establish robustness when operating images differ. ESA identifies the work as a master’s thesis collaboration and notes that it had not yet produced a peer-reviewed publication.

That finding reinforces a practical design requirement: efficiency and robustness need to be evaluated together. A smaller model that runs quickly but fails under realistic image degradation may offer less mission value than a slower alternative with better demonstrated behavior.

This emphasis should also be distinguished from proposals for large orbital computing facilities. New Space Economy’s examination of orbital AI workloads frames the commercial issue around whether a particular workload benefits from its location. Processing near an orbital sensor and building a general-purpose data center in space are different investment propositions.

For the roadmap, the immediate industrial implication is specialization. Businesses can contribute through embedded software, processing hardware, testing, fault recovery, model adaptation, and integration. They do not necessarily need to own a large constellation or develop a general-purpose foundation model to supply an important part of the system.

Assurance Is a Product Capability, Not a Final Check

Assurance is among the roadmap’s strongest recurring themes. It means developing justified confidence that a system will perform acceptably within its intended conditions and limits. For space AI, this requires more than showing that a model produces accurate outputs on a selected dataset.

Verification examines whether a system meets specified requirements. Validation examines whether it is appropriate for the intended use. Certification, where an applicable process exists, involves a formal determination against defined criteria. These concepts are connected, but they are not interchangeable, and none should be treated as a universal guarantee of safety.

The roadmap recommends space-specific benchmarks, uncertainty analysis, explainability tools, simulation environments, and methods for assessing autonomous systems. It also proposes incorporating AI validation and performance expectations into funding and procurement. These are recommendations for developing practice and capability, rather than evidence that a comprehensive space-AI certification regime already exists.

The broader AI Risk Management Framework developed by the U.S. National Institute of Standards and Technology provides a useful international comparison. It is voluntary and addresses trustworthiness throughout the design, development, use, and evaluation of AI systems. It is not a spacecraft approval mechanism or a substitute for mission-specific engineering evidence.

The roadmap’s commercial argument is that expertise in assurance could itself become a valuable service. Testing facilities, independent evaluation, qualified software components, and evidence packages could help operators decide which capabilities to adopt. A smaller supplier may find a defensible market by solving an assurance problem that larger mission teams also face.

There is a cost tradeoff. Demanding evidence can slow deployment and consume resources that would otherwise fund new features. Requirements should reflect the consequences of failure and the authority assigned to a system. An analytical aid reviewed by a specialist and a control function acting without immediate supervision should not automatically face identical assessment processes.

Assurance also continues after deployment. Changes to a model, its input data, its hardware, or its operating environment can affect the validity of earlier evidence. A credible service needs controlled updates, performance monitoring, and a way to restrict operation when confidence is insufficient.

The strategic implication is that trust must be supported by observable practice. International influence will depend on whether other organizations accept the evidence and find the methods useful. National ambition alone cannot establish a globally recognized standard or persuade a customer to rely on an autonomous capability.

Shared Data and Procurement Determine Whether Suppliers Can Scale

The roadmap’s seventh investment theme proposes a national space data and AI infrastructure layer. Its eighth recommendation adds a specialized coordination hub. Together, these proposals address fragmentation among organizations that may possess complementary skills but lack practical arrangements for sharing resources and progressing toward operational deployment.

Data access is a central issue. Developers need representative measurements to train, evaluate, and improve systems. Spacecraft telemetry, which records information about equipment and operations, can be commercially sensitive or restricted for security reasons. Unusual failures may also be poorly represented, making some of the most important situations difficult to study.

The roadmap proposes shared datasets, benchmarking environments, test facilities, and federated learning. Federated learning allows participating systems to contribute to model development without necessarily pooling all their raw data in one place. It can support collaboration, but it does not remove the need for agreements about access, security, intellectual property, and responsibility.

A national coordination function could help identify common needs and connect projects to mission opportunities. Its usefulness would depend on what it enables participants to do. Access to appropriate test data, shared evaluation methods, and a credible deployment route would offer more practical value than coordination measured only through meetings or organizational membership.

Procurement is equally important. A research grant can establish technical feasibility, but it does not automatically produce a customer, a support organization, or continuing revenue. The roadmap calls for mission-aligned investment and public-sector demand that can help promising services move beyond isolated pilots.

For implementation, that implies defining the service being purchased. Requirements might address detection performance, response times, availability, interoperability, and the handling of uncertainty. A supplier then has a clearer basis for deciding what to build and how to demonstrate that it meets the customer’s needs.

The workforce requirement extends beyond AI researchers. Space engineers must understand how learning systems behave, and data specialists must understand spacecraft constraints and mission operations. Procurement teams, domain experts, cybersecurity personnel, and assurance specialists also influence whether a service can be adopted responsibly.

Finance and contractual risk remain part of the same picture. Buyers and suppliers need workable arrangements for maintenance, updates, data rights, and responsibility for failures. Investors need evidence that technical progress can lead to lasting demand. The roadmap recognizes the importance of coordinated support, but its proposals still require concrete delivery arrangements.

The broader lesson is that scaling depends on reducing repeated integration and adoption costs. A strong prototype is a starting point. A repeatable product must also be supportable, evaluable, and compatible with the operational systems into which it will be introduced.

The 2036 Ambition Needs Evidence at Each Stage

The roadmap divides development into three horizons. The short term, 2026 through 2030, emphasizes foundations, demonstrations, early adoption, and emerging assurance arrangements. The medium term, 2031 through 2036, emphasizes wider integration, commercial growth, and international influence. The longer horizon, beginning in 2037, extends the vision toward sustained capability and strategic resilience.

These periods organize priorities rather than establish guaranteed delivery dates. The document explicitly explains that activities within its roadmap do not follow a fixed sequence and can begin or finish at different points within a timeframe. Reading every item as a dated government commitment would misrepresent the source.

The eight recommendations collectively cover national space awareness, PNT resilience, satellite communications, EO services, spacecraft autonomy, launch operations, shared infrastructure, and national coordination. Their breadth is useful because it shows that the proposed transition reaches beyond onboard software. It also creates a management challenge: progress across so many areas requires clear priorities and identifiable responsibility.

The roadmap suggests five dimensions for evaluating success: technical and mission performance, economic outcomes, sovereign resilience, ecosystem and workforce strength, and international influence. Proposed indicators include deployed capabilities, reduced manual intervention, lower onboard energy use, commercial revenue, improved threat detection, collaborative projects, and adoption of assurance approaches.

These are useful starting points, but they need baselines and definitions to become operational measures. A reduction in manual intervention is beneficial only if required performance and safety are maintained. Increased revenue needs a clear method for identifying the contribution attributable to AI. More collaborative projects do not necessarily demonstrate stronger commercial services.

The roadmap does not provide a quantified delivery target for every proposed indicator. Implementation would need to establish who collects the evidence, how often progress is assessed, and what happens when an approach fails to deliver. These arrangements would help distinguish learning from a pilot from simply extending it.

For countries outside the UK, the framework offers a relevant strategic pattern. A nation does not need to dominate every layer of the space economy to supply valuable capabilities. Specialized engineering, trusted analytical services, efficient computing, and internationally usable assurance can support participation in larger systems.

That opportunity remains conditional. Customers must value the service, partners must accept the evidence, and suppliers must sustain performance beyond demonstrations. The strongest measure of success in 2036 would be a collection of dependable capabilities that users continue to purchase and operate, supported by institutions able to maintain and improve them.

Summary

The UK’s AI roadmap for space presents a strategy for competing through dependable applications, efficient systems, and integration expertise. Its eight capability categories and seven investment themes connect research to services spanning Earth observation, communications, navigation, orbital security, spacecraft operations, and launch.

The document’s most useful contribution is its treatment of adoption as a combined technical, commercial, and institutional task. Better algorithms require representative data, suitable hardware, operational testing, skilled personnel, and procurement that gives suppliers a route to sustained demand. Assurance is central because the consequences of an automated decision depend on the mission and the authority assigned to the system.

Examples from European Earth observation missions and NASA’s Mars exploration demonstrate that bounded AI capabilities already have practical uses. They also show why individual successes should not be confused with general autonomy or readiness for every mission.

The roadmap’s 2036 ambitions remain conditional on implementation. Progress will be most convincing when customers receive measurable improvements, operators can demonstrate dependable behavior, and proposed national capabilities become supported services rather than a continuing series of pilots.

Appendix: Useful Books Available on Amazon

• Artificial Intelligence: A Modern Approach
• Human Compatible: Artificial Intelligence and the Problem of Control
• Prediction Machines, Updated and Expanded: The Simple Economics of Artificial Intelligence
• Power and Prediction: The Disruptive Economics of Artificial Intelligence
• The Ethical Algorithm: The Science of Socially Aware Algorithm Design

Appendix: Top Questions Answered in This Article

What Is the UK’s Artificial Intelligence Roadmap for Space?

It is a strategic framework commissioned by the UK Space Agency and published on October 27, 2026. It identifies AI capabilities, investment themes, and supporting arrangements intended to strengthen the UK space sector through 2036, with an additional longer-term horizon beginning in 2037.

Does the Roadmap Announce Funding for Every Recommendation?

No. The roadmap presents recommendations, development pathways, and suggested indicators rather than a funding commitment for every activity. Establishing a proposed capability would require appropriate implementation decisions, resources, and delivery arrangements. Publication alone does not demonstrate that a program has been procured or deployed.

Why Does the Roadmap Emphasize Resource Efficiency?

Spacecraft operate with constrained power, memory, computing capacity, and communications. Efficient models can make useful onboard processing possible within those limits. The roadmap also identifies this expertise as a potential UK advantage, particularly when efficiency is combined with reliable behavior and evidence appropriate to the mission.

How Can AI Improve Earth Observation Services?

AI can help classify imagery, detect changes, combine observations, and produce assessments that support decisions. The commercial value depends on accuracy, timeliness, uncertainty, and integration into customer workflows. Processing more data is not sufficient if the resulting information does not improve a useful operational or business outcome.

What Does Spacecraft Autonomy Actually Mean?

Autonomy means that a system can perform specified tasks without detailed human direction at every step. Its scope may be narrow, such as selecting a measurement target or adjusting a schedule. Authority, supervision, and failure consequences differ between tasks, so autonomy should always be described in terms of particular capabilities.

Can AI Prevent Satellite Navigation Disruption?

AI can contribute by identifying suspicious signals, detecting interference, and combining information from multiple sources. It cannot independently guarantee uninterrupted navigation or timing. Effective resilience also requires suitable equipment, alternative information, clear operating procedures, and a way to communicate when the available solution cannot be trusted.

Why Is Assurance So Important for Space AI?

Space AI may influence expensive assets and decisions with significant operational consequences. Assurance provides evidence about intended performance, limitations, and acceptable use. It includes testing, monitoring, controlled changes, and appropriate supervision, rather than relying only on a model’s accuracy under selected development conditions.

Is Edge AI the Same as an Orbital Data Center?

No. Edge AI places processing close to a data source, which may be a satellite instrument or spacecraft subsystem. An orbital data center is a broader computing infrastructure proposition. A small onboard model can provide mission value without requiring a large computing facility in space.

Why Do Shared Data and Procurement Matter?

Developers need representative data and test environments to establish that systems work under realistic conditions. Procurement supplies the connection between technical capability and a paying customer’s requirements. Without both, promising projects can remain isolated demonstrations that are difficult to integrate, support, or sell repeatedly.

How Should Progress Toward 2036 Be Measured?

Progress should combine operational performance with commercial and institutional outcomes. Relevant measures include dependable deployments, improved efficiency, recurring revenue, resilience, workforce capability, and accepted assurance methods. Each measure needs a baseline and clear definition so that increased activity is not mistaken for improved results.

Appendix: Glossary of Key Terms

Artificial Intelligence

A broad field concerned with computational systems that perform tasks such as recognizing patterns, making predictions, planning actions, or interpreting information. In space, individual applications usually have a defined purpose and operating scope rather than general authority over an entire mission.

Digital Twin

A digital model of a physical asset or system that can incorporate relevant measurements and operational information. Engineers can use it to examine behavior, compare possible actions, or investigate faults. Its usefulness depends on the model’s accuracy and suitability for the intended question.

Edge Computing

Processing performed close to the equipment or sensor producing the data, rather than sending all information to a distant computing facility. On a spacecraft, it can support prompt responses or reduce transmissions, subject to available power, memory, processing capacity, and reliability requirements.

Earth Observation

The collection of information about Earth’s surface, atmosphere, and related conditions using remote sensing instruments. Satellite observations can support environmental monitoring, agriculture, disaster assessment, and other activities. Turning measurements into useful products generally requires processing, interpretation, validation, and connection to a user’s needs.

Space Domain Awareness

Understanding the objects, activities, and conditions present in space and their implications for operations. It combines observations with analysis to support decisions. Tracking an object’s movement is one component, but interpreting behavior and assessing risk also require attention to uncertainty.

Positioning, Navigation, and Timing

Services and techniques used to determine location, guide movement, and maintain accurate time. Satellite signals are one important source. Resilience depends on recognizing unreliable information and maintaining suitable alternatives when signals are unavailable, disrupted, or deliberately manipulated.

In-Orbit Servicing, Assembly, and Manufacturing

Activities involving the maintenance, construction, or production of hardware in orbit. The term covers distinct capabilities with different requirements, including inspection, repair, assembly, and fabrication. Demonstrating one task does not establish that a system can safely perform every activity in the category.

Inference

The use of a trained AI model to produce an output from new input data. For a spacecraft, this could mean classifying an image or assessing equipment measurements. Inference is distinct from training, which develops or adjusts the model using data.

Assurance

The evidence and reasoning used to establish justified confidence that a system is suitable for a defined purpose. It can involve testing, analysis, monitoring, and controlled changes. Assurance describes supported confidence within stated limits, rather than a promise that failure is impossible.

Federated Learning

An approach to developing a model through contributions from multiple participating systems without necessarily collecting all their raw data in one central location. It can support collaboration, but still requires arrangements for security, data quality, access, ownership, and evaluation of the resulting model.