ReVAD: From Imitation to Reasoning in Vectorized Autonomous Driving via Latent Space Search

Zhao Yang, Chengkang Duan, Weiyi Hu, Haoran Hu, Hua Cui, Qingshuang Sun
Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, PMLR 337:7751-7760, 2026.

Abstract

Vectorized end-to-end (E2E) autonomous driving models predominantly rely on reactive imitation learning. By failing to model underlying environmental dynamics, these agents lack counterfactual reasoning capabilities and struggle with uncertainty in out-of-distribution ({OOD}) scenarios. To bridge the gap between reactive imitation and deliberative decision-making, we introduce Reasoning-enhanced VAD ({ReVAD}), empowering agents with a “System 2” cognitive framework via Latent Space Search. Unlike computationally expensive pixel-level world models, {ReVAD} integrates a probabilistic Token Dynamics Model with Latent Monte Carlo Tree Search to perform efficient lookahead planning entirely within a sparse semantic token space. By simulating future states to evaluate risk and utilizing imitation policies solely as search priors, {ReVAD} effectively mitigates the distribution shift inherent in pure imitation learning, demonstrating significantly improved robustness and safety in high-uncertainty environments.

Cite this Paper

BibTeX
@InProceedings{pmlr-v337-yang26b, title = {{ReVAD}: From Imitation to Reasoning in Vectorized Autonomous Driving via Latent Space Search}, author = {Yang, Zhao and Duan, Chengkang and Hu, Weiyi and Hu, Haoran and Cui, Hua and Sun, Qingshuang}, booktitle = {Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence}, pages = {7751--7760}, year = {2026}, editor = {Perković, Emilija and Malinsky, Daniel}, volume = {337}, series = {Proceedings of Machine Learning Research}, month = {17--21 Aug}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v337/main/assets/yang26b/yang26b.pdf}, url = {https://proceedings.mlr.press/v337/yang26b.html}, abstract = {Vectorized end-to-end (E2E) autonomous driving models predominantly rely on reactive imitation learning. By failing to model underlying environmental dynamics, these agents lack counterfactual reasoning capabilities and struggle with uncertainty in out-of-distribution ({OOD}) scenarios. To bridge the gap between reactive imitation and deliberative decision-making, we introduce Reasoning-enhanced VAD ({ReVAD}), empowering agents with a “System 2” cognitive framework via Latent Space Search. Unlike computationally expensive pixel-level world models, {ReVAD} integrates a probabilistic Token Dynamics Model with Latent Monte Carlo Tree Search to perform efficient lookahead planning entirely within a sparse semantic token space. By simulating future states to evaluate risk and utilizing imitation policies solely as search priors, {ReVAD} effectively mitigates the distribution shift inherent in pure imitation learning, demonstrating significantly improved robustness and safety in high-uncertainty environments.} }
Endnote
%0 Conference Paper %T ReVAD: From Imitation to Reasoning in Vectorized Autonomous Driving via Latent Space Search %A Zhao Yang %A Chengkang Duan %A Weiyi Hu %A Haoran Hu %A Hua Cui %A Qingshuang Sun %B Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Emilija Perković %E Daniel Malinsky %F pmlr-v337-yang26b %I PMLR %P 7751--7760 %U https://proceedings.mlr.press/v337/yang26b.html %V 337 %X Vectorized end-to-end (E2E) autonomous driving models predominantly rely on reactive imitation learning. By failing to model underlying environmental dynamics, these agents lack counterfactual reasoning capabilities and struggle with uncertainty in out-of-distribution ({OOD}) scenarios. To bridge the gap between reactive imitation and deliberative decision-making, we introduce Reasoning-enhanced VAD ({ReVAD}), empowering agents with a “System 2” cognitive framework via Latent Space Search. Unlike computationally expensive pixel-level world models, {ReVAD} integrates a probabilistic Token Dynamics Model with Latent Monte Carlo Tree Search to perform efficient lookahead planning entirely within a sparse semantic token space. By simulating future states to evaluate risk and utilizing imitation policies solely as search priors, {ReVAD} effectively mitigates the distribution shift inherent in pure imitation learning, demonstrating significantly improved robustness and safety in high-uncertainty environments.
APA
Yang, Z., Duan, C., Hu, W., Hu, H., Cui, H. & Sun, Q.. (2026). ReVAD: From Imitation to Reasoning in Vectorized Autonomous Driving via Latent Space Search. Proceedings of the 42nd Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research 337:7751-7760 Available from https://proceedings.mlr.press/v337/yang26b.html.

Related Material