{"id":"75e9dffc-773c-4ef7-b0a0-1de2a0f9a850","slug":"robostral-navigate","name":"Robostral Navigate","description":"Robostral Navigate is Mistral AI's first model for embodied navigation. An 8B parameter model that takes RGB images and natural language instructions to guide robots through complex environments. State-of-the-art on R2R-CE (76.6% success on validation unseen, 79.4% on validation seen). Operates with a single RGB camera, no LiDAR or depth sensors. Built entirely in-house, trained in simulation on ~400K trajectories across 6,000 scenes. Generalizes across wheeled, legged, and flying robots.","brainType":"foundation-model","isOpen":false,"maturityStage":"research","architecture":"8B parameter VLM initialized from vision-language model for grounding (pointing, counting, object localization). Navigation via pointing (infers image coordinates of target location) with fallback to metric displacements. Trained with prefix-caching (tree-based attention masking, 22x token reduction) and online RL (CISPO algorithm, +3.2% improvement).","reviewStatus":"unreviewed","sources":[{"url":"https://mistral.ai/news/robostral-navigate/","title":"Robostral Navigate: single-camera AI navigation (Mistral AI official)","sourceName":"mistral.ai","publishedAt":"2026-07-08"},{"url":"https://x.com/MistralAI/status/2075278815417528448","title":"Mistral AI on X: Announcing Robostral Navigate, 8B robotics navigation model, SOTA on R2R-CE","sourceName":"X"},{"url":"https://jobs.lever.co/mistral?team=Research","title":"Mistral AI robotics team hiring page","sourceName":"mistral.ai"}],"keyFacts":[{"label":"Type","value":"Foundation model (embodied navigation)"},{"label":"Size","value":"8B parameters"},{"label":"Input","value":"RGB images + natural language instructions"},{"label":"Sensors required","value":"Single RGB camera only; no LiDAR or depth sensors"},{"label":"Benchmark","value":"State-of-the-art on R2R-CE: 76.6% success on validation unseen, 79.4% on validation seen"},{"label":"Margin","value":"Beats best single-camera approach by 9.7 points; beats best multi-sensor by 4.5 points despite no depth sensors"},{"label":"Training","value":"Built entirely in-house; trained in simulation on ~400K trajectories across 6,000 scenes; no reliance on existing open-source VLMs"},{"label":"Efficiency","value":"Prefix-caching with tree-based attention masking compresses episode into single sequence; 22x token reduction vs per-timestep training"},{"label":"RL post-training","value":"Online reinforcement learning (CISPO algorithm) improved success rate by 3.2% after supervised training"},{"label":"Generalization","value":"Runs on wheeled, legged, and flying robots; generalizes across robot sizes and camera intrinsics"},{"label":"Navigation method","value":"Pointing-based (infers target coordinates in camera view) with fallback to metric displacements when target out of view"},{"label":"Authors","value":"Theo Cachet, Arjun Majumdar, Srijan Mishra, Thomas Chabal, Chris Bamford, Elliot Chane-Sane, Benjamin Tibi, Ludovic Ho Fuh, Olivier Duchenne (AI Science Robotics)"}],"aliases":["Robostral Navigate","Robostral"],"collisionRisk":"low","reviewNote":null,"builtOnBrainId":null,"createdAt":"2026-07-10T06:32:01.094Z","updatedAt":"2026-07-10T06:32:01.094Z","jsonLd":{"@context":"https://schema.org","@type":"SoftwareApplication","@id":"https://registry.deploy.report/brains/robostral-navigate","url":"https://registry.deploy.report/brains/robostral-navigate","name":"Robostral Navigate","alternateName":["Robostral Navigate","Robostral"],"description":"Robostral Navigate is Mistral AI's first model for embodied navigation. An 8B parameter model that takes RGB images and natural language instructions to guide robots through complex environments. State-of-the-art on R2R-CE (76.6% success on validation unseen, 79.4% on validation seen). Operates with a single RGB camera, no LiDAR or depth sensors. Built entirely in-house, trained in simulation on ~400K trajectories across 6,000 scenes. Generalizes across wheeled, legged, and flying robots.","identifier":"75e9dffc-773c-4ef7-b0a0-1de2a0f9a850","applicationCategory":"foundation-model","publisher":{"@id":"https://deploy.report/#organization"}},"framework_metadata":{"framework_schema_version":"0.1.0","verification_status":"unreviewed","maturity_stage":"research","lifecycle_state":null,"architectural_position":{"cohort":null,"sub_cohorts":[]},"within_cohort_verified_vs_claimed_pair":null,"cap_flags":[],"verification_depth":{"sources_count":3,"primary_source_types":[]}}}