Mistral launches its first robotics model, expanding into physical AI

Paris-based Mistral AI has officially entered the "Physical AI" market, unveiling its first-ever robotics model tailored for industrial automation and autonomous navigation.

Mistral AI on Wednesday unveiled its robotics model as Europe's leading AI company pushes into factories, warehouses and industrial automation.

The launch follows Mistral's acquisition of Austria's Emmi AI in May and comes months after Paris-based startup Genesis AI unveiled a broader robotics model with navigation and manipulation capabilities.

Dubbed Robostral Navigate, the 8-billion-parameter (8B) model signals a major strategic pivot for Europe's leading generative AI startup.

Physical AI enable machines to interact with users for operating in the real world.

Unlike traditional AI models that predominately process text, images or speech, physical AI allows robots and autonomous systems to perceive their surroundings, make decisions and perform physical tasks.

Key features

This specific model has been specially designed to work with robots from different suppliers.

Its main feature enables robot navigation using a single camera and the system does not require lidar, advanced sensors or multiple-camera setups.

Moving beyond purely digital language and code generation, Mistral is pushing directly into physical environments, targeting factories, warehouses, logistics, and hospitality.

It mainly focuses on navigation rather than object handling or manipulation.

The model combines AI software with sensors, cameras, actuators and robotics hardware to enable machines to move, manipulate objects and respond to changing environments.

The company is developing AI that can serve as the "brain" of robots, helping them understand instructions, plan actions and adapt to real-world conditions.

Such models can be used in industrial robots, warehouse automation, manufacturing, logistics and even household robots.

Despite using vastly less sensing hardware, the model achieved a 76.6% success rate on unseen R2R-CE benchmarks (the industry standard for robots following instructions in unfamiliar settings)-outperforming the previous best single-camera approach by 9.7 points.

The move reflects a broader industry trend as AI companies are increasingly looking beyond chatbots and text-generation tools to develop models that can control physical devices.