SimVTA: Simulation-Grounded Video-Trajectory Alignment for Driving World Action Models