Ego-Dynamics-Augmented World Model for Autonomous Driving with Zero-Shot Cross-Chassis Adaptation

2026-07-14

AI Summary

This work addresses a critical limitation in existing world model–based autonomous driving approaches, wherein the ego-vehicle’s motion is conflated with scene dynamics in bird’s-eye-view representations, degrading modeling fidelity and imagination accuracy. To resolve this, the authors propose a dynamics-augmented Dreamer-style reinforcement learning framework that explicitly incorporates ego-vehicle dynamics as a prior. By encoding the ego-vehicle’s state history into a disentangled contextual representation and modulating the latent distribution of a causal Transformer-based world model, the method achieves decoupled modeling of vehicle dynamics and environmental motion. This design enables zero-shot transfer across vehicle chassis without retraining. Experimental results demonstrate substantial improvements, with task success rates increasing by 28% in urban settings and 61% on highways, alongside a 73% performance advantage in zero-shot transfer to unseen chassis configurations.

Abstract

World model (WM)-based reinforcement learning enables sample-efficient end-to-end autonomous driving learning by imagining long-horizon trajectories in latent space. However, most driving WMs operate on bird's-eye-view (BEV) representations that are inherently egocentric: the transition between consecutive frames entangles the ego vehicle's own motion with scene dynamics. As a result, the WM devotes significant capacity to recovering ego-motion from warped observations, at the cost of scene modeling fidelity and imagination accuracy. This work proposes DynaDreamer, a dynamics-augmented Dreamer-style reinforcement learning method to address this problem by augmenting the WM with an explicit ego-dynamics prior. A physics-informed ego-dynamics encoder-decoder extracts the ego-state history into a compact and identifiable context, which modulates a causal Transformer WM to condition both its prior and posterior latents. During imagination, the ego-dynamics predictor propagates this context forward to keep the ego-dynamics prior synchronized with the rollout. An information-theoretic analysis shows that conditioning on this context reduces both the predictive entropy of the observation transition and the prior--posterior Kullback--Leibler divergence, confining the WM's modeling burden to the scene dynamics beyond ego-motion. An additional benefit is zero-shot cross-chassis adaptation: the ego-dynamics context depends on identifiable chassis parameters, so that a vehicle with previously unseen dynamic characteristics can adapt the WM to the new chassis without retraining. Experiments demonstrate that DynaDreamer improves task success rates over the strongest baseline by 28% and 61% in urban and highway driving scenarios, respectively, with the advantage rising to 73% when extrapolating to unseen chassis.

Problem

• world model
• ego-dynamics
• autonomous driving
• cross-chassis adaptation
• scene dynamics

Innovation

• ego-dynamics prior
• world model
• zero-shot cross-chassis adaptation
• causal Transformer
• reinforcement learning