VITAL: From Visual Motion to Precise Action
VITAL’s one-step driving model nearly matches its ten-step score at 3.29× the end-to-end inference speed.
The paper reports 90.8 PDMS on NAVSIM v1 navtest, within 0.13 points of its ten-step result. VITAL pairs a motion scaffold derived from visual movement with a compact correction for the trajectory errors that scaffold leaves behind. At inference, it predicts both from the current image, vehicle state and navigation command, without generating future images. Without retraining, it scores 91.1 EPDMS on NAVSIM v2 navtest and 38.6 on navhard.
HF Daily Papers' note
The paper reports 90.8 PDMS on NAVSIM v1 navtest, within 0.13 points of its ten-step result. VITAL pairs a motion scaffold derived from visual movement with a compact correction for the trajectory errors that scaffold leaves behind. At inference, it predicts both from the current image, vehicle state and navigation command, without generating future images. Without retraining, it scores 91.1 EPDMS on NAVSIM v2 navtest and 38.6 on navhard.
HF Daily Papers' note
score 4