DriveMA: Driving Vision-Language-Action Models with verifiable Meta-Actions Paper • 2605.31271 • Published 13 days ago