VaViM and VaVAM: Autonomous Driving through Video Generative Modeling

TL;DR

This paper introduces VaViM and VaVAM, large-scale generative video models for autonomous driving, demonstrating how video pre-training improves scene understanding and trajectory generation in driving scenarios.

Contribution

The paper presents novel auto-regressive video and video-action models that leverage large-scale pre-training for autonomous driving tasks.

Findings

Video models capture driving scene semantics and dynamics.

Pre-training enhances perception and planning in driving.

Scaling models improves video synthesis and understanding.

Abstract

We explore the potential of large-scale generative video models for autonomous driving, introducing an open-source auto-regressive video model (VaViM) and its companion video-action model (VaVAM) to investigate how video pre-training transfers to real-world driving. VaViM is a simple auto-regressive video model that predicts frames using spatio-temporal token sequences. We show that it captures the semantics and dynamics of driving scenes. VaVAM, the video-action model, leverages the learned representations of VaViM to generate driving trajectories through imitation learning. Together, the models form a complete perception-to-action pipeline. We evaluate our models in open- and closed-loop driving scenarios, revealing that video-based pre-training holds promise for autonomous driving. Key insights include the semantic richness of the learned representations, the benefits of scaling for video synthesis, and the complex relationship between model size, data, and safety metrics in closed-loop evaluations. We release code and model weights at https://github.com/valeoai/VideoActionModel