xiaomi-research/recogdrive
ReCogDrive is a research framework that combines vision-language models, diffusion policies, and reinforcement learning for end-to-end autonomous driving.
Velocity · 7d
+1.5
★ / day
Trend
→steady
star history
ReCogDrive is an ICLR 2026 paper presenting a cognitive framework for autonomous driving that integrates vision-language models with diffusion-based action policies and reinforced learning. The system enables end-to-end autonomous navigation from raw sensory inputs using a NavSim-based evaluation pipeline. The repository provides model weights on HuggingFace along with a pretraining dataset for the framework.