Sequence of Expert: Boosting Imitation Planners for Autonomous Driving through Temporal Alternation

ICLR 2026 Conference Submission17415 Authors

Keywords: Imitation learning, Autonomous driving

TL;DR: This work proposes Sequence of Expert, a plug-and-play method to improve performance of IL planners. Through extensive evaluation, we show that SoE is universally effective for all IL planners.

Abstract: Imitation learning (IL) has emerged as a central paradigm in autonomous driving. While IL excels in matching expert behavior in open-loop settings by minimizing per-step prediction errors, its performance degrades unexpectedly in closed-loop due to the gradual accumulation of small, often imperceptible errors over time. Over successive planning cycles, these errors compound, potentially resulting in severe failures. Current research efforts predominantly rely on increasingly sophisticated network architectures or high-fidelity training datasets to enhance the robustness of IL planners against error accumulation, focusing on the state-level robustness at a single time point. However, autonomous driving is inherently a continuous-time process, and leveraging the temporal scale to enhance robustness may provide a new perspective for addressing this issue. To this end, we propose a method termed Sequence of Experts (SoE)—a temporal alternation policy that enhances closed-loop performance without increasing model size or data requirements. The key idea is to retain intermediate models from training that possess inherent differences in driving errors, and then alternate the activation of different models at certain temporal intervals. This approach not only preserves the consistency capability across multiple models but also leverages their differences to enhance robustness. As a plug-and-play solution for existing IL planners, our approach requires no architectural modifications or prior knowledge of scenarios, making it highly practical for real-world deployment. Our experiments on large-scale autonomous driving benchmarks nuPlan demonstrate that SoE method consistently and significantly improves the performance of all the evaluated models, and achieves state-of-the-art performance. This module may provide a key and widely applicable support for improving the training efficiency of autonomous driving models.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Submission Number: 17415