Tenders Are Invited For Qualified Software Machine Learning Toolkit For Space Hardware - Expro Plus

Tenders are invited for Qualified Software Machine Learning Toolkit for Space Hardware - Expro plus. The objective of this activity is to develop and qualify a software toolkit for machine learning deployment and inference in space hardware.Description: Artificial Intelligence (AI), especially in the form of Machine Learning (ML), has made its way to the space domain in recent years. It is promising to revolutionise the future of space applications by enabling higher level on-board autonomy and new on-board features not previously possible. However, despite the recent efforts in the field, the deployment of ML in space hardware remains challenging for different reasons: Space qualified (radiation hardened) hardware impose limitations in terms of performance compared to terrestrial equivalents. Software qualification requirements for institutional missions imposes limitations for the used underlying operating systems and libraries. Monolithic toolchains, including hardware specific compilers, are required for some hardware. This makes it not easily possible to reuse optimization techniques across different devices and creates the need to maintain different toolchains for different targets. Currently, no fully qualified hardware/software solution exists for automatic deployment of machine learning applications in space. This activity aims to answer these challenges and contribute to the rapid adoption of AI and ML in space applications and increase the reuse of libraries and lower the associated costs related to qualification of custom software.The activity will develop a software toolkit, specifically targeting space hardware and space qualification according to the relevant ECSS standards. The toolkit may target both space qualified classical symmetric multiprocessing (SMP) multicore processors (e.g. SPARC, RISC-V, ARM), and processors with hardware acceleration for machine learning inference (e.g. implemented in FPGA logic).The toolkit will allow for flexibility, by ensuring compliance with existing machine learning training frameworks and standard model formats (such as the open ONNX format). The toolkit shall also allow for modular expansion, for additional software (operating systems) and space hardware. The following tasks are foreseen in the frame of this activity: Identify at least two different space representative hardware targets to be covered by the toolkit. Identify at least two relevant use cases, making use of AI, to be use as demonstrators. Derive the user requirements and the technical requirements for the toolkit including the definition of requirement related to model optimization, deployment and inference. Define a workflow to extend the toolkit to additional software/hardware targets. ESA UNCLASSIFIED - For ESA Official Use Only ESA/IPC(2024)61,add.6 Annex II Page 5/10 Design and implement the end-to-end toolkit. Perform software qualification according to the relevant ECSS standards. Deliver a demonstrator on at least two different space representative hardware targets. Deliverables: Qualified software; Report. Tender Link : https://esastar-publication-ext.sso.esa.int/ESATenderActions/filter/open