Enabling satellite operators to offer AI at the edge in space

This post has been republished via RSS; it originally appeared at: Microsoft Tech Community - Latest Blogs - .



Azure Space fuses cloud computing and space technology, empowering industries and innovators with connectivity and access to space data to reach for the stars. Azure is committed to shaping the future of the aerospace industry with cutting-edge solutions. We look forward to connecting with global space leaders at the 39th Space Symposium in Colardo Springs. Please email us to request a meeting, and join us for our speaking sessions:


The Power of Processing From Space: In-orbit innovations

Wednesday, April 10, 2024 | 12:45 PM MT | Cheyenne Mountain Resort
Steve Kitay, Senior Director of Azure Space joins industry experts for this panel discussion highlighting disruptive technologies that are making waves in the market.


Generative AI: Changing the way we approach content creation

Thursday April 11, 2024 | 8:45 AM MT | International Center South
Angel Smith, General Manager of Azure Mission Engineering will moderate a panel on Generative AI. Angel will moderate a panel of industry experts on Generative AI.


Exploring the possibilities for AI on-orbit

Over the past 2 years Microsoft has been working with industry leaders to explore leveraging AI on satellites to detect, identify and optimize insights for earth observation missions. In March 2024 we saw the launches of some of our first on-orbit testbeds and are validating and exploring new cases with academia, commercial and government customers with our partners:


Thales Alenia

The IMAGIN-e mission is unlocking on-orbit compute for Earth observation insights immediately applicable for our planet’s surveillance and protection, enabling data scientists to explore moving AI models from the cloud to the edge while learning how to optimize the balance between them. We have opened up this platform as part of an international challenge launched by the European Space Agency’s Φ-lab to foster creativity and kickstart the next generation of geospatial research.


Loft Orbital

YAM-6 is the first in a new class of spacecraft, designed to support Virtual Missions. Azure Orbital Space edge SDK is a solution that enables the deployment of Virtual Missions on Loft Infrastructure. Virtual Missions enable governments and enterprises to run AI based applications at the edge, generating real-time insights directly from space-based sensors. Customers for Virtual Missions include Space Compass, who is working with Loft and Microsoft to run a virtual mission on YAM-6 to demonstrate AI-based ship detection.


Building out our on-orbit testbed platforms

Computing power in space has made giant strides in the last few years. Taking advantage of these new capabilities requires developing new software for satellites, incorporating learnings from IOT and cloud. Traditional satellite software development requires specialized skills, tools, and methodologies, which poses challenges for both existing and new entrants in the market. Compared to the broader enterprise application development market, which has a large and diverse pool of developers, the space industry has a much smaller and more specialized pool of software engineers, who need to have both domain knowledge and technical skills for developing satellite software. This talent gap can hamper the innovation and competitiveness of the satellite operators, especially as new entrants and startups enter the market with disruptive technologies and business models.


Diagram showing the Azure Space SDK in the context of app development in the cloud vs app execution on-orbit.Diagram showing the Azure Space SDK in the context of app development in the cloud vs app execution on-orbit.


Microsoft is enabling any developer to be a space developer by reducing the barriers to entry for the space industry. Enterprise developers can leverage familiar tools and languages, such as VS code, C# and Python. The Azure Orbital Space edge SDK is creating an ecosystem of satellite providers that leverage open standards, so that any satellite operator can enable AI applications on their satellites simply and securely. Based on our research and collaboration with our partners we have incorporated these core design considerations:


Payload compute separate from safety critical compute

The fundamental computing operations for spacecraft health and safety are different than those needed to extract and process insights. Typically, low-level health and safety software is run on radiation hardened, but less computationally powerful compute nodes. De-coupling AI and image processing from these flight computers have benefits in terms of security isolation and being able to leverage a more powerful consumer grade compute nodes with GPUs, but lower radiation tolerance and flight heritage.


Multi-architecture and multi-node compute with Kubernetes

As spacecraft designers are looking at the compute capabilities, they are evaluating characteristics that are specific to their overall mission objectives. A pattern is emerging of leveraging multiple compute nodes together, with different nodes having special capabilities such as a GPU or being run at different security isolation levels. Kubernetes provides a common abstraction and orchestration of the computing resources, regardless of their location, type, or configuration.


Spacecraft abstraction through an extendable Host Platform

Each spacecraft has different capabilities, sensors, and communication approaches, yet a lot of the core Earth observation (EO) tasks that need to be achieved are common regardless of satellite configuration. We leverage the same platform for all our on-orbit testbeds and place the per spacecraft specific code on our extensible ‘plug-in’ Host Platform. The Host Platform exposes a set of common API’s that enable application developers to focus on higher insight and decision-making code, leveraging cutting edge AI libraries rather than low-level space-craft interactions to obtain and process data.


Developing in the cloud with virtual satellites and extendable data generators

Having access to space-craft hardware is difficult during development on the ground – and even harder once the satellite is operational in orbit. Virtual satellites in Azure, enable developers to build and test applications through extendable data generators that mimic satellite sensors. One of the most common issues we’ve encountered is testing against representative imagery data. To facilitate this, we’ve created data generators that tie into the Microsoft Planetary Computer which hosts petabytes of free satellite data.


Learn More

Leave a Reply

Your email address will not be published. Required fields are marked *


This site uses Akismet to reduce spam. Learn how your comment data is processed.