Valeo and KE:SAI launch open-source Franco-German push for generalist self-driving AI
Valeo and the open-science lab KE:SAI will jointly build a generalist autonomous driving AI system and release all of its code, trained models and datasets as open source.
- €20.9 billion
- 100,000
- 149
- May 2026
What Happened
Valeo, the French automotive technology company, and KE:SAI (Kyutai ELLIS Scalable Autonomous Intelligence), the non-profit open-science research lab co-founded by kyutai and the ELLIS Institute Tübingen, announced an open-science research collaboration on October 13, 2026, in Paris. The two will jointly develop and openly release models, code and weights for autonomous driving research, aiming at a fully reproducible, generalist AI driving system for autonomous vehicles. Findings will be shared openly through publications, and suppliers and researchers will be able to reproduce the results, build on the model and benchmark their own systems against a common reference.
- Every component — code, trained models and datasets — released as open source
- A fully reproducible, generalist AI driving system for autonomous vehicles
- Performance that first matches, then pushes beyond, state-of-the-art results across all major public driving benchmarks
- Open publications, so suppliers and researchers can reproduce results and benchmark their own systems against a common reference
“This open-science collaboration with KE:SAI marks a strategic milestone for Valeo and the broader European automotive ecosystem. By combining our deep industrial and technological expertise in autonomous driving with world-class open-source research, we are establishing a shared foundation to accelerate the development of physical AI and democratize next-generation self-driving technologies.”
“For language models, open source compressed years of progress into months, yet frontier physical AI is still largely closed. Valeo shares our belief that open foundations create more innovation than any closed system ever could. Everything we develop will be open. Every step forward is a step forward for everyone.”
KE:SAI is founded by kyutai and the ELLIS Institute Tübingen, with a founding donation from kyutai, and is co-located in Tübingen and Paris.
Valeo and KE:SAI announce in Paris an open-science collaboration to develop and openly release a generalist AI driving system.
The initiative joins two European teams: valeo.ai, Valeo's AI research lab in Paris, whose open-source models and publications such as DrivoR and VaVAM are widely used in the community, and the KE:SAI founding team in Tübingen, known for open benchmarks and driving stacks such as KITTI, NAVSIM and TransFuser. Their stated ambition is to strengthen the European and worldwide automotive ecosystems by giving automakers, suppliers and research labs a built-in-Europe, open and shared foundation for autonomous driving systems. The partners argue that physical AI is widely considered the next frontier of artificial intelligence, yet leading efforts, including in self-driving, remain largely closed, with nearly every driving model deployed in the industry proprietary. Without a shared foundation, they say, automakers and suppliers depend on technology they cannot inspect, adapt or reproduce, and open questions on efficiency and generalization cannot be pursued by a single lab — developing in the open lets researchers across institutions test, challenge and build on the work and brings academic and industrial research closer together.
On the technical side, the partnership aims to build one compact AI model that can drive different vehicles, with different sensors, in different environments. The teams will begin by training on large-scale real-world driving recordings, with a model designed to transfer across different sensor and camera configurations. From there, the model is meant to go beyond what recorded driving alone can teach: through reinforcement learning in simulation and learned world models, it will be exposed to rare and safety-critical situations that are scarce or absent in real-world data.
Previously from Valeo
Valeo has also been building its position in automated driving systems with carmakers: in a story we published on October 1, 2026, Nissan picked Valeo as its system Tier-1 supplier for the carmaker's next-generation ProPILOT program. In that role, Valeo integrates Wayve AI Driver software into an automotive-grade hardware and software platform.
Background drawn from MotorClaw's earlier coverage of Valeo's official releases.
Why this matters
Nearly every driving model deployed in the automotive industry today is proprietary, so automakers, suppliers and researchers have no shared, open foundation they can inspect, adapt or reproduce. The partners say releasing code, models and datasets openly will give European and global automakers, suppliers and research labs a common reference to build on and benchmark against, and speed up progress in physical AI.
Terms in This Story
- Physical AI
- Artificial intelligence that perceives and acts in the physical world, as robots and self-driving vehicles must.
- World models
- AI models that learn how an environment changes over time so they can predict future situations and plan actions.
- Reinforcement learning
- A training method in which a model learns by trial and error, improving on the actions that lead to better outcomes.
- Weights
- The trained parameters inside an AI model; releasing them openly lets others run, study and adapt the model.
Summarised from the linked release; details can be imperfect — always verify against the original source.