Reproduce “Driver-WM: A Driver-Centric Traffic-Conditioned Latent World Model for In-Cabin Dynamics Rollout” 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
Haozhuang Chi, Daosheng Qiu, Hao Su, Haochen Liu, Zirui Li, Haoruo Zhang, Chen Lv
This research introduces Driver-WM, a driver-centric latent world model developed at Nanyang Technological University, which predicts in-cabin driver kinematics and semantics over multiple steps, explicitly conditioned on external traffic conditions. The model improves long-horizon kinematic fidelity in reactive driving scenarios and enhances semantic consistency for driver behavior and emotion compared to prior approaches.
Key insights
- Explicitly modeling the causal link from external traffic to internal driver dynamics via a gated injection mechanism is crucial for accurate and anticipatory predictions of human reactions.
- Operating in a compact, semantically rich latent space derived from Vision-Language Models enables efficient and stable long-horizon rollouts of structured driver states (kinematics and semantics).
- Test-time interventions on the gated causal injection mechanism provide a valuable diagnostic tool to analyze how specific external cues influence driver responses without physical experimentation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.