Reproduce “HanoiWorld : A Joint Embedding Predictive Architecture BasedWorld Model for Autonomous Vehicle Controller” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Tran Tien Dat, Nguyen Hai An, Nguyen Khanh Viet Dung, Nguyen Duy Duc

Researchers at Hanoi University of Science and Technology developed HanoiWorld, a world model for autonomous vehicle controllers that integrates a pre-trained V-JEPA 2 encoder with a Recurrent State-Space Model. The system achieves lower collision rates in highway and roundabout scenarios compared to baselines, while also demonstrating a significant failure mode in merging environments.

Key insights

  • Integrating a pre-trained V-JEPA 2 encoder significantly enhances representation learning for autonomous driving, effectively extracting robust, decision-relevant features while reducing computational burden compared to pixel-based methods.
  • Combining JEPA-based encoding with an RSSM allows for efficient long-term planning and decision-making within a compact latent space, minimizing reliance on extensive real-world data.
  • The world model can achieve improved safety performance, specifically lower collision rates, in specific driving scenarios by focusing on predictive representation learning.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.