World-action models for autonomous driving typically train action experts on trajectory representations that are defined independently of visual dynamics. The authors of VITAL first show that reconstructing trajectories from recorded future images captures the dominant maneuver but leaves centimeter-scale errors, motivating a representation that ties action targets directly to visual motion.

VITAL (Vision-Induced Tokens for Action Learning) couples a visually derived motion scaffold with a compact residual that encodes the trajectory error left by that scaffold. This makes visual motion the principal prediction target while giving small metric corrections a separate, normalized learning scale. Using flow matching, the action expert jointly predicts both components from the current image, ego state, and navigation command—without generating future images at inference.

With only one action denoising step, VITAL achieves 90.8 PDMS on NAVSIM v1 navtest, within 0.13 points of its ten-step score at 3.29x the end-to-end inference speed. Without retraining, it reaches 91.1 EPDMS on NAVSIM v2 navtest and 38.6 on navhard. The same residual coordinates also support bounded reinforcement post-training that refines execution while keeping the predicted visual scaffold fixed.