KE:SAI Open Science Autonomy Lab (kesai.eu)

KE:SAI, a new non-profit AI research lab co-founded by kyutai and the ELLIS Institute Tübingen, has launched with a mission to advance physical AI research through open and reproducible practices. The lab aims to address the current bottlenecks in robot learning caused by the high costs of physical interaction and the reliance on proprietary, closed systems. By embracing an open-source framework, KE:SAI seeks to democratize this field and foster collaboration among researchers, aiming to accelerate innovation and reduce the fragmentation that currently characterizes the space. Significantly, KE:SAI plans to develop data- and compute-efficient methods for training foundational models that apply hybrid, causal, and latent world models. Unlike traditional data-driven approaches, these models will incorporate diverse data sources, including real-world information and simulations. The lab’s initial focus will be on self-driving technologies, offering a promising avenue to demonstrate their capabilities—achieving competitive performance with lower data and computational requirements. By aligning with the open science ethos, KE:SAI represents a noteworthy step towards making physical AI research more accessible and collaborative, ultimately enhancing its potential for real-world applications.