HMVLM: Multistage Reasoning-Enhanced Vision-Language Model for Long-Tailed Driving Scenarios

TL;DR

HMVLM is a multistage vision-language driving model that enhances long-tailed scenario handling through multi-stage reasoning, history-aware prompting, and trajectory smoothing, achieving top performance in a major autonomous driving challenge.

Contribution

The paper introduces HMVLM, a novel multistage reasoning-enhanced vision-language model with specific prompting and post-processing techniques for improved autonomous driving in complex scenarios.

Findings

Achieved second place in the 2025 Waymo Vision-based E2E Driving Challenge.

Surpassed the public baseline by 2.77% in Rater Feedback Score.

Effectively handles long-tailed driving scenarios with multi-stage reasoning.

Abstract

We present HaoMo Vision-Language Model (HMVLM), an end-to-end driving framework that implements the slow branch of a cognitively inspired fast-slow architecture. A fast controller outputs low-level steering, throttle, and brake commands, while a slow planner-a large vision-language model-generates high-level intents such as "yield to pedestrian" or "merge after the truck" without compromising latency. HMVLM introduces three upgrades: (1) selective five-view prompting with an embedded 4s history of ego kinematics, (2) multi-stage chain-of-thought (CoT) prompting that enforces a Scene Understanding -> Driving Decision -> Trajectory Inference reasoning flow, and (3) spline-based trajectory post-processing that removes late-stage jitter and sharp turns. Trained on the Waymo Open Dataset, these upgrades enable HMVLM to achieve a Rater Feedback Score (RFS) of 7.7367, securing 2nd place in the 2025 Waymo Vision-based End-to-End (E2E) Driving Challenge and surpassing the public baseline by 2.77%.