Valeo Partnered With KE:SAI for Open-Source Driving AI
The collaboration aims to create a public, reproducible reference model to compete with opaque, proprietary automotive software.
Updated on Oct. 13, 2026 in Artificial Intelligence
Live Poll
Is the development of open-source artificial intelligence better for technological progress than proprietary systems?
Valeo and the non-profit lab KE:SAI announced a partnership to build an open-source, generalist AI system for autonomous driving. The project will release all code, trained models, and datasets publicly to provide a transparent reference for the industry.
Why it matters
Most automakers currently rely on proprietary driving models that lack transparency and reproducibility. This initiative aims to establish a public benchmark for AI safety and capability that researchers can examine and modify.
The team aims to match and surpass existing scores on the KITTI and NAVSIM benchmarks using a single compact model. This will be evaluated against proprietary stacks that currently dominate automotive sensor data processing.
The players
Valeo
A global automotive supplier that operates the valeo.ai research laboratory and develops advanced driver-assistance systems.
KE:SAI
A non-profit research laboratory founded by Kyutai and the ELLIS Institute Tübingen focused on developing open-source artificial intelligence.
The details
The project combines the research expertise of valeo.ai with the team from the ELLIS Institute Tübingen. The system is trained on real-world driving data using simulation-based reinforcement learning—a machine learning method where agents learn by trial-and-error in a digital environment—to encounter rare, safety-critical events. By developing learned world models, the system predicts environmental dynamics, allowing it to adapt across varied vehicles and sensor configurations.
Timeline
October 13, 2026: The partnership between Valeo and KE:SAI was formally announced.
The Tech Race
The project utilizes the KITTI and NAVSIM benchmarks as the primary performance metrics for its open-source driving model. It follows an established research pattern of using these standards to challenge proprietary automotive stacks.
This project aims to standardize how driving AI is validated, potentially shifting industry reliance toward public, auditable software. While the model is currently in development, its future release will allow developers to test and build on these tools for diverse sensor sets and vehicles.
The takeaway
The partnership marks a shift toward transparency in the autonomous driving sector by prioritizing public model reproducibility over closed-source stacks. Industry participants should monitor upcoming releases of the team's trained models and their comparative scores on the KITTI and NAVSIM benchmarks.
Further reading
For broader trends in open-source development, visit our Artificial Intelligence section.
Live Poll
Is the development of open-source artificial intelligence better for technological progress than proprietary systems?
