Mistral AI has introduced Robostral Navigate, its first robotics navigation model, as the Paris-based company seeks to expand its presence in the physical AI sector covering industrial automation, manufacturing, and logistics.
Robostral Navigate is designed to guide robots through complex and dynamic environments using only a single RGB camera and simple language instructions.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
According to Mistral AI, the model achieves a 76.6% success rate on the room-to-room in continuous environments (R2R-CE) benchmark, which measures how well an agent can follow instructions in environments not seen during training.
Mistral AI states this result outperforms existing models using either a single camera or multiple sensors such as depth cameras and LiDAR.
The 8bn parameter model operates by processing images together with plain-language commands, enabling autonomous movement through a range of settings such as offices, factories, warehouses, and public spaces.
Unlike previous solutions that often require additional hardware, Robostral Navigate relies solely on visual data from a standard camera.
The system has been built entirely in-house by Mistral AI and trained using simulated data in order to generalise across different types and sizes of robots.
Robostral Navigate has been tested on various robot platforms, including wheeled, legged, and aerial vehicles. The model is said to remain effective even when encountering obstacles or scenarios not included in its training dataset.
The AI framework uses a pointing-based navigation method, predicting target locations as image coordinates in the robot’s current field of view.
When the target is not visible, the system falls back to standard displacement commands based on the robot’s local frame.
The robot has potential uses in sectors such as manufacturing, delivery, logistics, and hospitality, as it is able to interpret and follow complex, multi-step instructions without prior exposure to a specific environment.
Mistral AI’s recent advances follow a period of notable investor interest. The company has been in discussions to raise approximately €3bn ($3.5bn) at a possible valuation of up to €20bn.
In September last year, Mistral AI reached a valuation of €11.7bn, with ASML taking an 11% stake for €1.3bn.
Founded in 2023 by researchers previously at Google DeepMind and Meta Platforms, Mistral AI is positioning itself as a European player in a field largely dominated by firms from the US and China.
The company is also developing cloud-computing infrastructure to support European government and business clients, with facilities planned in France and Sweden.
