EponaV2: Driving World Model with Comprehensive Future Reasoning

TL;DR

EponaV2 is a novel driving world model that forecasts comprehensive future scene representations, enhancing real-world reasoning and trajectory planning without relying on manual annotations.

Contribution

It introduces a new paradigm that predicts detailed future geometry and semantics, inspired by human anticipation, and employs a flow matching policy optimization for improved planning.

Findings

Achieves state-of-the-art performance on NAVSIM benchmarks.

Improves trajectory planning accuracy by forecasting comprehensive future scene data.

Demonstrates effective perception-free driving world modeling.

Abstract

Data scaling plays a pivotal role in the pursuit of general intelligence. However, the prevailing perception-planning paradigm in autonomous driving relies heavily on expensive manual annotations to supervise trajectory planning, which severely limits its scalability. Conversely, although existing perception-free driving world models achieve impressive driving performance, their real-world reasoning ability for planning is solely built on next frame image forecasting. Due to the lack of enough supervision, these models often struggle with comprehensive scene understanding, resulting in unsatisfactory trajectory planning. In this paper, we propose EponaV2, a novel paradigm of driving world models, which achieves high-quality planning with comprehensive future reasoning. Inspired by how human drivers anticipate 3D geometry and semantics, we train our model to forecast more comprehensive future representations, which can be additionally decoded to future geometry and semantic maps. Extracting the 3D and semantic modalities enables our model to deeply understand the surrounding environment, and the future prediction task significantly enhances the real-world reasoning capabilities of EponaV2, ultimately leading to improved trajectory planning. Moreover, inspired by the training recipe of Large Language Models (LLMs), we introduce a flow matching group relative policy optimization mechanism to further improve planning accuracy. The state-of-the-art (SOTA) performances of EponaV2 among perception-free models on three NAVSIM benchmarks (+1.3PDMS, +5.5EPDMS) demonstrate the effectiveness of our methods.