PriorEye: Geospatial Visual Priors for End-to-End Autonomous Driving (ECCV 2026)
rsasaki0109 · x · 2026-08-23
Addressing the limitation of end-to-end autonomous driving methods that rely solely on instantaneous sensor observations, this paper (ECCV 2026) introduces PriorEye. It leverages geospatial visual priors—street-level visual context anchored to the planned route—to provide visual-spatial foresight similar to human driver experience.
Technical Contributions:
- Dual-Memory Architecture: Combines a contextual memory for retrieved priors with a persistent fallback memory.
- Adaptive Memory Gate: Dynamically regulates the influence of memories based on compatibility with the current driving state.
- Seamless Integration: The memory augmentation module is designed to be easily integrated into existing end-to-end approaches.
Performance:
Evaluated on the NAVSIM-v2 benchmark, the approach demonstrates significant improvements in driving performance by incorporating anticipatory prior knowledge.
More from Embodied
- Realbotix: the Las Vegas company betting on robots that look very much like humans — kyliebytes · 2026-10-16
- Stanford's Fei-Fei Li team open-sources OpenWAM, a unified world-action model framework for robots — jiqizhixin · 2026-10-16
- Robot folding leaderboard: GPT-6 Astra leads at 1349 Elo, Muse Wonder trails at 1034 — qinzytech · 2026-10-16
- NVIDIA's TacitVLA: robots reason without reading their own thoughts, 3.4x lower latency — arankomatsuzaki · 2026-10-16
- SEAR: first large-scale controlled study of self-evolving robot agents, 7 LLMs, 330 tasks — xiye_nlp · 2026-10-16
- Podcast: subsea robotics startup Ulysses on operating 1,500m underwater and ocean carbon — blakemenezes · 2026-10-16