Its Robostral Navigate AI model needs input from just one color camera, doing without Lidar, depth sensors, or multiple viewpoints.
French AI company Mistral claims its latest AI model offers a more efficient way to train and operate robots.
The model, Robostral Navigate, can guide a robot through plain language instructions, using a single RGB camera to find its way. Mistral said that this was a radical departure from most other models, which rely on depth sensors, LiDAR or several cameras working together.
Robostral Navigate has achieved a score of 76.6% on the R2R-CE (Room-to-Room in Continuous Environments) benchmark for robots following instructions. This beats the best system using depth sensors or multiple cameras by 4.5 percentage-points, despite the Robostral Navigate using neither of these aids, and puts it 9.7 percentage-points ahead of the next-best single-camera robot.
Mistral said it had designed the model to autonomously navigate complex environments including offices, residential and commercial buildings, and outdoor settings. A key feature of the new model is that it is easier to train: Mistral said the number of training tokens is reduced significantly compared to other models, reducing training runs from months to days.
Robotics is an area ripe for AI research: The World Economic Forum at Davos in February heard how AI-driven robotics could drive advances in productivity.
Other AI model developers are ahead of the game: Nvidia announced robotic AI efforts in August 2025.
This article first appeared on Computerworld.
