Valeo and KE:SAI are working on an open AI driving system, with plans to release its code, trained models and datasets. The project aims to give automakers and researchers a reproducible benchmark in a field where most production driving models are proprietary.
Watch Desk analysis
What happened
Automotive World reports that Valeo’s Paris research lab, valeo.ai, is partnering with KE:SAI, a non-profit lab founded by Kyutai and the ELLIS Institute Tübingen. The partners want to build one compact model that can handle different vehicles, sensor sets and settings, and plan to match then beat leading scores on public driving benchmarks.
Training is due to start with real-world driving data, followed by simulation-based reinforcement learning and learned world models to expose the system to rare, safety-critical situations that recorded data may miss. The Automotive World report says the code, models and datasets are intended to be released openly.
Why it matters
Driving software is increasingly central to vehicle development, but proprietary systems can leave manufacturers dependent on technology they cannot inspect or adapt. An open reference model could give automakers, suppliers and researchers something concrete to test, compare and build on.
The ambition is notable; the proof will be in the results. Beating public benchmarks would be a useful milestone, but real-world driving spans messy conditions that a leaderboard cannot capture on its own.
Our read
This is a substantial bet on openness in physical AI, where safety and varied hardware make reproducibility especially valuable. The promise is not that open code makes autonomous driving safe by magic. It is that more people may be able to scrutinise the system and measure how it performs. We’ll be interested to see what the partners publish, and how they test it beyond benchmark scores.
What to watch
- When the partners release the code, trained models and datasets.
- Which public benchmarks they choose, and whether they publish comparable results.
- How the system performs across different vehicles, sensors and safety-critical scenarios.
Discussion spark: Should an open driving model be judged mainly by public benchmark scores, or should real-world testing and safety evidence be the deciding test?
Sources and evidence
- Source update (13 October 2026, 04:40 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.