Driving · 2026-10-20

From Likelihood to Utility: Quality-Axis Diffusion for End-to-End Autonomous Driving

Junjie Yang, Ziyi Shi, Shuqi Shen, Xinhu Zheng, Hongliang Lu, Hai Yang

AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.

TL;DR

QualiDrive replaces diffusion’s Gaussian-noise coordinate with a driving-quality axis for end-to-end autonomous driving. It builds scene-specific quality ladders by ranking observed trajectories using measured driving utility, then learns reverse transitions for ordered low-to-high-quality refinement. On NAVSIM, the authors report 94.5 PDMS (v1) and 92.74 EPDMS (v2).

Why it matters

The authors argue that conventional diffusion progresses toward likely demonstrations rather than higher driving utility. By embedding measured quality directly into the diffusion coordinate and learning ordered refinement steps, QualiDrive aims to produce candidate pools with denser high-quality trajectories, reducing reliance on downstream selection.

Method

Constructs scene-specific intent-preserving “quality ladders” by ranking a dense trajectory library with measured driving utility and enforcing non-regressing ordered transitions.
Trains a quality-axis diffusion model (QualiDrive) whose reverse steps condition on scene, ego state, current trajectory, and target quality level, refining trajectories from low to high quality.
Uses a multi-objective scorer (nine sub-scores) to rank generated candidates; selection picks the highest composite estimated score.

Limitation

The authors state evaluation is limited to non-reactive NAVSIM and evaluator-derived supervision; reactive closed-loop behavior, independent safety or preference judgments, and out-of-distribution generalization remain important future tests.

Abstract (from arXiv)

Diffusion models have become a dominant paradigm for end-to-end autonomous driving, yet existing planners organize trajectory generation along the Gaussian-noise axis, treating driving utility only as an auxiliary guidance or selection signal. We instead formulate diffusion along a driving-quality axis. We propose QualiDrive, which constructs scene-specific quality ladders by ranking observed trajectories according to measured driving utility and learns to reverse this ordered degradation through ordered low-to-high-quality refinement. Each reverse step is conditioned on the scene, ego state, current trajectory, and target quality level, progressively refining trajectories toward higher driving quality. On NAVSIM, QualiDrive achieves 94.5 PDMS on v1 and a state-of-the-art 92.74 EPDMS on v2. QualiDrive improves Oracle EPDMS only modestly over endpoint-only quality supervision (98.52 to 98.57), but substantially increases the fractions of candidates above 0.95 and 0.98 EPDMS from 12.94%/9.71% to 21.53%/15.93%, raising selected EPDMS from 92.63 to 92.74. These results suggest that diffusion is not merely a process of denoising, but one of progressive quality-ordered refinement.

Publication record

Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.

First public release: 2026-09-26 · Indexed by Pub-AI: 2026-10-20 UTC

preprint