Mistral’s 8B Robostral Navigate uses one RGB camera and tops R2R-CE at 77.4%
mistralai · hf · 2026-07-24
What Mistral released
Robostral Navigate is an 8B vision-language model for robot navigation that uses only a single monocular RGB stream to predict the next waypoint in image space.
Why it matters
The model is designed to avoid depth sensors, multi-camera rigs, and prebuilt maps, making deployment easier across different robot embodiments.
Training recipe
- 2.4 million trajectories
- 350k simulated scenes
- A prefix-caching recipe that packs full episodes into single sequences
- Training tokens reduced by 22x, cutting training time from months to days
- A tree-based attention mask to prevent leakage from ground-truth actions
- Reinforcement learning to improve exploration and recovery
Results
- R2R-CE: 77.4% success rate, +10.5 points over the best monocular method and +5.3 points over the strongest depth- or multi-camera system
- RxR-CE: 75.1% success rate, beating all monocular baselines
Takeaway
Mistral is pushing robot navigation toward a cheaper, single-camera, cross-embodiment deployment model.
More from Embodied
- RealSense D585 Pro AI-Powered 3D Stereo Depth Camera Unboxed for Physical AI — chrismatthieu · 2026-10-26
- Physical world agent swarm demo shared on X — pduan · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26
- 'Lamp Swarm': a desk of lamp-shaped robots shows what physical AI work looks like — dee_hw · 2026-10-26
- Tencent's HiSTEVE 1.0: hierarchical VLA agent plays Minecraft with native keyboard and mouse, hits 85% SOTA — _akhaliq · 2026-10-26
- SJTU's VaM lets robots learn online from mistakes, 95%+ success over 4-hour autonomous runs — chris_j_paxton · 2026-10-26