Search papers, labs, and topics across Lattice.
This paper introduces DynaDreamer, a dynamics-augmented reinforcement learning method that enhances world models for autonomous driving by incorporating an explicit ego-dynamics prior. By using a physics-informed encoder-decoder to extract and synchronize ego-state history, the model significantly improves scene modeling fidelity and imagination accuracy while reducing the burden of ego-motion recovery. Experimental results show that DynaDreamer achieves a 28% and 61% increase in task success rates in urban and highway driving scenarios, respectively, and a remarkable 73% improvement when adapting to unseen chassis configurations without retraining.
Zero-shot adaptation to new vehicle dynamics leads to a 73% boost in autonomous driving success rates without retraining.
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.
