Bench-2Drive

The 'Bench-2Drive' benchmark is used to evaluate the performance of Vision-Language Action models in autonomous driving, specifically assessing their efficiency and planning quality through experiments that measure transformer-layer sparsity and latency reduction.

Papers using Bench-2Drive (8)

DriveStack-VLA: Render-Teacher Alignment for BEV-Based DeepStack Vision-Language-Action Model2026Orion-Lite: Distilling LLM Reasoning into Efficient Vision-Only Driving Models2026Judge, Then Drive: A Critic-Centric Vision Language Action Framework for Autonomous Driving2026Drive My Way: Preference Alignment of Vision-Language-Action Model for Personalized Driving2026MindDriver: Introducing Progressive Multimodal Reasoning for Autonomous Driving2026CoT4AD: A Vision-Language-Action Model with Explicit Chain-of-Thought Reasoning for Autonomous Driving2025Deead: Dynamic Early Exit Of Vision-language Action For Efficient Autonomous Driving2025X-driver: Explainable Autonomous Driving With Vision-language Models2025