Introducing Robostral Navigate

Robostral Navigate is our SOTA vision-language-action model built for embodied navigation. Give it a plain-language instruction and it moves a robot through any space. No LiDAR, no depth sensors, no multi-camera rig. State of the art on R2R-CE. One 8B model, trained entirely in simulation, running on wheeled, legged, and flying robots and generalizing across robot sizes. Navigation is the foundation for general-purpose robotics. Talk to our team: mistral.ai