KE:SAI

KE:SAI (Kyutai ELLIS Scalable Autonomous Intelligence) is a non-profit frontier AI research lab co-founded by kyutai and the ELLIS Institute Tübingen, co-located in Tübingen and Paris. It targets researchers and engineers in physical AI who are frustrated by the closed, proprietary silos that dominate robot learning and autonomous driving. The lab's mission is to advance the efficiency frontier of robust and safe physical AI through fully open and reproducible research, democratizing access to foundation models and training stacks that are currently locked behind corporate walls. KE:SAI is a small, focused team of world-leading researchers—with 27 best paper awards at top vision, learning, and robotics conferences—committed to releasing all code and models under permissive licenses allowing civil commercial use.

KE:SAI develops efficient and open foundation models using hybrid, causal, and latent world models. Unlike current data-driven imitation learning approaches, the lab pioneers Sim2Real techniques that integrate synthetic data with real-world data, enabling policies to generalize across a wide variety of embodiments and environments. The hybrid world models ingest information from real-world data, simulations, and common-sense knowledge from existing foundation models. Expressive, causally grounded latent spaces allow data- and compute-efficient closed-loop training of robot policies using (self-)supervised learning, reinforcement learning, and self-play objectives. The lab will first demonstrate these capabilities with a fully open self-driving stack, training policies using significantly less data and compute than typical while targeting infraction rates competitive with frontier systems.

In the physical AI research landscape, KE:SAI positions itself as a non-profit open-science alternative to proprietary labs. Its competitors include Infinite Blue, Exiger, and Ideagen, but KE:SAI differentiates by its exclusive focus on open and reproducible research, its hybrid causal world model approach, and its commitment to releasing a complete self-driving stack. The lab's founders introduced the KITTI benchmarks (over 10 million downloads), achieved top places at international self-driving competitions (nuPlan, CARLA, Waymo), and open-sourced the two largest driving datasets (OpenDV and PhysicalAI-AV, each 1700+ hours). This track record gives KE:SAI credibility that few other open-science initiatives can match.

The honest trade-offs: KE:SAI is a non-profit lab with no specified pricing or commercial support, meaning users cannot rely on SLAs or enterprise-grade deployment assistance. The lab's focus on causal world models and Sim2Real is ambitious but unproven at scale compared to established data-driven methods. As a small team, KE:SAI may produce fewer benchmarks and models than larger corporate labs. Finally, the fully open self-driving stack is a first demonstration—extensions to other robotics domains remain future work, not current deliverables.

Not specified in the provided sources; as a non-profit lab, KE:SAI releases all code and models under permissive licenses.