$ cat wiki/models/robostral-navigate.md
Robostral Navigate
modelupdated 2026-07-21created 2026-07-09
Spec
| Attribute | Value |
|---|---|
| Developer | Mistral AI |
| Released | 2026-07-08 |
| Announced | 2026-07-08 |
| Context window | unknown |
| Pricing | unknown |
| License | TBD (not open-weight in initial release) |
| Availability | unknown |
| Parameters | 8B |
| Sensors | single RGB camera (no depth sensors, no LiDAR required) |
| Input | current camera image + natural-language navigation instruction |
| Output | pointing — image coordinates of the target location + desired orientation on arrival |
| Training environment | simulation only (no real-robot data required) |
| RL algorithm | CISPO (online RL; +3.2 percentage points success-rate improvement) |
| Deployment | hardware-agnostic (any robot platform) |
Release Date
2026-07-08. Announced via Mistral blog (source) and covered by Bloomberg.
Benchmarks
| Benchmark | Robostral Navigate | Prior best (single-camera) | Prior best (depth/multi-cam) |
|---|---|---|---|
| R2R-CE (Room-to-Room in Continuous Environments) — validation unseen | 76.6% | ~66.9% | ~72.1% |
| Outperforms all prior navigation systems including those using depth sensors or multiple cameras — despite using neither. |
Method
Robostral Navigate predicts robot movement via pointing: given the current camera frame, it outputs the pixel coordinates of the target destination and the desired approach orientation. This is robust to changes in camera intrinsics and world scale, unlike metric-displacement command approaches. Training is entirely simulation-based, using the CISPO online RL algorithm for policy improvement.
Use Cases
Autonomous robot navigation in complex physical environments:
- Manufacturing (assembly line navigation, floor robots)
- Delivery (last-mile indoor/warehouse robots)
- Logistics (warehouse navigation)
- Hospitality (front-of-house service robots)
Significance
Robostral Navigate is Mistral's first robotics model — a significant expansion from the language and coding domain into physical AI. Key points:
- Sim-only training closes the sim-to-real gap: achieving state-of-the-art R2R-CE with purely simulation training suggests the field's sim-to-real transfer problem is closer to solved than previously believed.
- Single-camera constraint as a feature: most deployed robots already have RGB cameras but not depth or LiDAR. A SOTA model requiring only a camera dramatically lowers the hardware bar for adoption.
- 8B class: small enough to run on-device (robot edge compute), unlike 7B+ VLM-based approaches requiring cloud inference.
- Hardware-agnostic: the pointing-based interface separates perception from actuation, making the model portable across robot hardware.
Compared To
| Model | Org | Sensors | R2R-CE | Notes |
|---|---|---|---|---|
| Robostral Navigate | Mistral | RGB only | 76.6% | New SOTA; sim-only training |
| Gemini Robotics ER 1.6 | Google DeepMind | Multi-sensor | — | Industrial inspection (gauge-reading 93%); different task |
| Cosmos 3 Super | NVIDIA | Multi-modal | — | World model / generalist; different approach |
Sources
- Mistral blog — Robostral Navigate
- Bloomberg — Mistral robotics push
- [sources/blogs/mistral-2026-07-08-robostral-navigate.md]
Related
- Mistral AI — developer
- Embodied Agents — physical AI context
- Gemini Robotics ER 1.6 — competing robotics model (DeepMind)
- Cosmos 3 Super — NVIDIA open physical AI model