DriveMA: Rethinking Language Interfaces in Driving VLAs with One-Step Meta-Actions
A research paper proposes DriveMA, a new approach for driving vision-language-action models (VLAs) that replaces verbose natural-language reasoning with concise one-step meta-actions. This method aims to overcome bottlenecks in annotation, model complexity, and inference latency. DriveMA achieved state-of-the-art results on the Waymo End-to-End Driving Challenge with both 2B and 4B parameter models, outperforming previous methods. AI
IMPACT Introduces a more efficient interface for driving AI, potentially improving real-world autonomous driving systems.