Vega: Learning to Drive with Natural Language Instructions

TL;DR

Vega is a unified vision-language-world-action model that enables personalized autonomous driving by following diverse natural language instructions, utilizing a large-scale dataset and advanced multimodal processing techniques.

Contribution

The paper introduces Vega, a novel multimodal model that integrates vision, language, and world modeling for instruction-based driving, along with a large annotated dataset.

Findings

Vega achieves superior planning performance.

The model exhibits strong instruction-following abilities.

Extensive experiments validate the effectiveness of the approach.

Abstract

Vision-language-action models have reshaped autonomous driving to incorporate languages into the decision-making process. However, most existing pipelines only utilize the language modality for scene descriptions or reasoning and lack the flexibility to follow diverse user instructions for personalized driving. To address this, we first construct a large-scale driving dataset (InstructScene) containing around 100,000 scenes annotated with diverse driving instructions with the corresponding trajectories. We then propose a unified Vision-Language-World-Action model, Vega, for instruction-based generation and planning. We employ the autoregressive paradigm to process visual inputs (vision) and language instructions (language) and the diffusion paradigm to generate future predictions (world modeling) and trajectories (action). We perform joint attention to enable interactions between the modalities and use individual projection layers for different modalities for more capabilities. Extensive experiments demonstrate that our method not only achieves superior planning performance but also exhibits strong instruction-following abilities, paving the way for more intelligent and personalized driving systems.