Reproduce “Sequence of Expert: Boosting Imitation Planners for Autonomous Driving through Temporal Alternation” 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
Xiang Li, Gang Liu, Weitao Zhou, Hongyi Zhu, Zhong Cao
The Sequence of Experts (SoE) framework improves imitation learning planners for autonomous driving by temporally alternating between models trained with different random seeds. This method significantly enhances closed-loop driving performance on the nuPlan benchmark without increasing inference latency or requiring additional training data.
Key insights
- Models trained with identical configurations but different random seeds develop complementary closed-loop performance characteristics, despite similar open-loop accuracies.
- Optimal closed-loop driving performance sometimes emerges from earlier training epochs, suggesting that 'over-training' for open-loop precision can be detrimental to robustness.
- Leveraging the temporal dimension and inter-model diversity through strategic alternation offers an efficient method to enhance robustness without escalating computational resources.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.