Ahead of the upcoming Eclipse SDV Community Days in Bonn, Germany, we spoke with Pierre Gaillard about the Eclipse Aidge project and how it can benefit the world of software-defined vehicles.
What problem does Eclipse Aidge aim to solve for edge AI in software-defined vehicles?
Eclipse Aidge tackles the core challenges of edge AI in software-defined vehicles by addressing hardware fragmentation and resource constraints. Vehicles rely on highly heterogeneous platforms (CPU, GPU, NPU, DSP, FPGA, and increasingly chiplet-based architectures), which forces developers to rewrite deployment pipelines for each target, while AI models often exceed embedded limits in memory, energy, and latency. Aidge provides a generic, hardware-agnostic framework combined with automated optimisation, enabling efficient deployment across heterogeneous targets and supporting hardware-software co-design so that models and architectures can be jointly optimised to fit automotive constraints without sacrificing accuracy.
Why is an open source approach important for edge AI platforms like Eclipse Aidge?
An open source approach is essential to ensure interoperability, collaboration, and trust: Eclipse Aidge allows OEMs and suppliers to build on a shared foundation “below the value line”, reducing redundant development and fostering ecosystem-wide innovation. Eclipse Aidge additionally provides open, traceable, and certification features, which are critical for safety-related functions.
“Eclipse Aidge provides a generic, hardware-agnostic framework combined with automated optimisation.”
How does Eclipse Aidge support running AI workloads at the vehicle edge rather than in the cloud?
Eclipse Aidge enables efficient on-vehicle inference through an end-to-end deployment pipeline that applies advanced optimisation techniques – including quantisation, model compression, and pruning – to adapt trained models to strict automotive constraints in terms of latency, memory footprint, and power consumption. Beyond model optimisation, Aidge provides flexible generation of lightweight source code, making it suitable for deeply embedded environments. It also produces hardware-aware implementations optimised for various hardware targets by leveraging specialised libraries, enabling AI workloads to run locally in the vehicle.
What kinds of use cases in software-defined vehicles could benefit from Eclipse Aidge?
Eclipse Aidge benefits a wide range of software-defined vehicle use cases, including ADAS and autonomous perception, driver and occupant monitoring, emergency braking and collision avoidance, in-cabin intelligence, and predictive maintenance, all of which demand high reliability, and efficient execution on resource-constrained, heterogeneous platforms.
In his talk at the Eclipse SDV Community Days in Bonn, Pierre will demonstrate how open-source embedded AI frameworks can contribute to building interoperable, sustainable, and SDV-ready software platforms.
Pierre Gaillard holds a PhD in machine learning from the University of Technology of Compiègne, as well as an engineering degree and a master’s degree in computer science and information systems. With over ten years of experience in data-driven innovation, he strives to bridge the gap between advanced data processing, artificial intelligence, and concrete industrial applications. Currently a project manager at the CEA, he leads the DeepGreen project, an initiative to develop Eclipse Aidge, an open-source platform dedicated to embedded AI.
