Robostral Navigate
#5 in Robotik & Embodied AIMistral AI · 15× · tolest 20. Juli 2026
50
Momentum
Robostral Navigate is the first robotics navigation model from Mistral AI, an 8-billion-parameter model for embodied navigation. It controls robots using only a single RGB camera image and a natural language instruction, without LiDAR, depth sensors, or multi-camera systems. On the R2R-CE Benchmark (Room-to-Room in Continuous Environments), it achieves a 76.6% success rate on unseen environments, outperforming both single-camera and multi-sensor approaches. The model was trained entirely in-house exclusively on simulated data and can be deployed hardware-agnostically on wheeled, legged, and aerial robots.
Momentum-Verloop
07.05.05.08.
Features
| Deployment Model | Trained entirely in simulation (~400,000 trajectories across 6,000 scenes); currently enterprise sales/pilot deployments, model weights reportedly not yet publicly released |
| Use Case Scope | Robotic navigation in offices, residential and commercial buildings, and outdoor settings; applications in manufacturing, delivery, logistics, and hospitality |
| Integrations | Accepts RGB images and text instructions; outputs pointing coordinates or local displacement commands for robot control |
| Platform | Hardware-agnostic: deployable on wheeled, legged, and flying robots using a single RGB camera; no LiDAR/depth sensors required |
| Release Date | July 8, 2026 |