Building on its experience in fostering transformative innovations aimed at market adoption in the Earth observation (EO) sector, the model of Φ-lab@ESRIN is being replicated across Europe by ESA. Under the leadership of the newly established Directorate of Commercialisation, Industry and Procurement, and Directorate of Earth Observation Programme, ESA has signed a Letter of Intent with AI Sweden to set up a pilot development of Φ-lab@Sweden (more information), a collaboration that will develop solutions in the area of Edge Computing and Learning in Space. Work will be closely coordinated with the Swedish Space Data Lab (Swedish Space Agency, Luleå University of Technology, RISE and AI Sweden).
The initiative will build on the success of the existing Φ-lab located in ESRIN in Frascati (Italy) and of InCubed EO commercialisation programme focusing on research which leads to insights or space-based solutions in the Earth Observation domain and is the first pilot of an ESA Φ-labNET as part of the new commercialisation strategy outlined in the ESA Agenda 2025.
Why edge computing?
Today, on Earth only 20% of data processing and analysis takes place locally, 80% in centralised data centres and computing facilities. While for now, the cloud business model is dominating (with a share of 75% of the European cloud market taken by non-EU players), it is predicted that in the next 5 years, most of the data processing and analytics will run at the edge of the network. A shift from general-purpose clouds to sector or task-specific applications might go hand in hand with the rise of edge computing.
Increasingly existing cloud networks face physical limitations: more data needs, but limited bandwidth and hence increasing costs. To deal with requirements for real-time access to data and large amounts of data being produced, part of the computation shall be moved to the “edge” of the cloud. Edge computing takes data processing closer to where the data is being collected to get near-real-time results and derive value from quickly acting on the data. With local processing power, data analysis and processing can be carried out locally, extracting actionable information that can be transmitted and further processed by other nodes of the networks. In addition, edge computing nodes can learn directly from local data how to extract such information thanks to the novel edge learning algorithms.
Across sectors, the way we monitor and control our environment will be changed by edge computing. It will impact IT infrastructure itself, but also applications in electromobility and autonomous vehicles, precision farming, or manufacturing. As the technology is sector agnostic, new possibilities for technology transfers and multiply spill over effects are being created. As an open ecosystem, even more innovation can be triggered. This decentralisation through new IoT and edge computing capabilities is becoming a chance for Europe to build up innovation-driven solutions. But how to benefit from this paradigm shift and how to develop next-generation IoT and edge computing infrastructure in space?
Edge computing and Edge learning in space
Colocation of computing resources, virtually running applications in the cloud, and edge computing are all familiar concepts to the data world. With increasing digitalisation, the space industry will further benefit from integrating these ideas into space-based business models. It can enable shorter download times to Earth and in general improve the agility and autonomy of the space infrastructure, hence saving time, money, and improving results. Edge devices on spacecraft process sensor data (e.g., images, gases, sampling), often leveraging machine learning models to produce actionable information on the spot. Edge Computing in Space is an emerging technology that can open a new domain for software applications, probably similar in scale to mobile apps. Few examples of main added value to space:
Examples of exploration use cases:
- Reducing the dependence on Earth and the time to get results for decisions or experiments (like zero-gravity experiments or DNA Sequencing),
- Improving crew health and facilities monitoring, particularly in preparation for deep-space exploration,
- Opening doors and facilitating new explorations (e.g., reducing data need and increasing data quality of rover activities).
Examples of satellite use cases:
- Bringing data centre services into Earth orbit and beyond is an emerging trend, by this increasing autonomy and resilience in decentralised networks,
- Rapid extraction of information onboard Earth Observation (EO) satellites: possible use cases for AI and satellite data include the identification of flooding, illegal fishing, deforestation, algal blooms, or environmental disasters,
- Onboard pre-filtering or discarding cloud-covered or corrupted satellite images. A preliminary demonstration was performed in the frame of the Ф-sat-1 mission, enabling “smart” discarding of cloudy images.
The Intel-Movidius Myriad 2 board (right) mounted on the top of HyperScout-2 electronics stack (left) is an example of edge-computing platform used onboard Ф-sat-1 satellite to detect cloud-covered images (Credits: https://directory.eoportal.org/web/eoportal/satellite-missions/p/phisat-1).
More information about a variety of AI4EO applications capitalizing on embedded AI and enabling cloud computing in space is available at: https://philab.phi.esa.int/.
Business focus and ESA’s engagement
Companies focusing on satellite-based edge networks are already raising private capital to upscale activities. The next years will most probably show more performance computing infrastructure in space to process, clean, and aggregate data from multiple sources. Advancements on radiation shielding and thermal management systems will be needed to allow the usage of more and more off-the-shelf processors in space environments. While several companies are already working on related products and services (e.g. edge infrastructure as a service), technical and marketing challenges remain.
The Φ-lab in ESRIN has already supported on-orbit AI & Edge Computing in the past. AI Sweden and ESA will further collaborate on current research resources, funding and infrastructure. Read more about this collaboration, edge learning in space and how to get involved.