Institution profile

North University of China

Academic institutionasia · cn
Official website
Research library7linked papers
Opportunities0open roles
Selected work

Representative Papers

Not All History Helps: Velocity-Aware Selective Memory for Long-Horizon End-to-End Autonomous Driving

Aug 16, 2026

This study addresses the challenges of unreliable historical states and motion evolution in long-horizon planning for end-to-end autonomous driving by proposing StableDrive. The method leverages Mamba operators to construct a selective momentum memory that enhances the robustness of historical representations, while introducing a motion-stage training scaffold to guide the model in perceiving dynamic evolution, thereby enabling efficient single-model deployment without ensembling. Experiments demonstrate that StableDrive achieves state-of-the-art performance on benchmarks such as nuScenes, reducing collision rates by 23.3% and attaining the highest EPDMS score on NAVSIM v2. These results indicate significant improvements in both safety and temporal consistency for long-horizon planning, validating the effectiveness of integrating structured memory mechanisms with stage-aware training in complex driving scenarios.

0 citationsRead paper

Robust Multi-View Classification under Noisy Supervision via Global Anchor Consensus

Jul 20, 2026

This work addresses the significant performance degradation of multi-view classification under high label noise, where supervision signals become unreliable. To mitigate this issue, the authors propose the Global Anchor Consensus mechanism (GALA), which introduces per-class global anchors shared across views as stable references. By measuring the distances between samples and both their assigned-class and competing-class anchors, and integrating classifier confidence to compute cross-view scrutiny scores, GALA adaptively reweights suspicious samples and corrects their labels. This approach enables noise-robust representation learning and consistently outperforms eight state-of-the-art methods across six benchmark datasets, demonstrating particularly strong performance under high noise rates and validating its effectiveness and robustness.

0 citationsRead paper
Recent publications

Latest Papers

Not All History Helps: Velocity-Aware Selective Memory for Long-Horizon End-to-End Autonomous Driving

Aug 16, 2026

This study addresses the challenges of unreliable historical states and motion evolution in long-horizon planning for end-to-end autonomous driving by proposing StableDrive. The method leverages Mamba operators to construct a selective momentum memory that enhances the robustness of historical representations, while introducing a motion-stage training scaffold to guide the model in perceiving dynamic evolution, thereby enabling efficient single-model deployment without ensembling. Experiments demonstrate that StableDrive achieves state-of-the-art performance on benchmarks such as nuScenes, reducing collision rates by 23.3% and attaining the highest EPDMS score on NAVSIM v2. These results indicate significant improvements in both safety and temporal consistency for long-horizon planning, validating the effectiveness of integrating structured memory mechanisms with stage-aware training in complex driving scenarios.

0 citationsRead paper

Robust Multi-View Classification under Noisy Supervision via Global Anchor Consensus

Jul 20, 2026

This work addresses the significant performance degradation of multi-view classification under high label noise, where supervision signals become unreliable. To mitigate this issue, the authors propose the Global Anchor Consensus mechanism (GALA), which introduces per-class global anchors shared across views as stable references. By measuring the distances between samples and both their assigned-class and competing-class anchors, and integrating classifier confidence to compute cross-view scrutiny scores, GALA adaptively reweights suspicious samples and corrects their labels. This approach enables noise-robust representation learning and consistently outperforms eight state-of-the-art methods across six benchmark datasets, demonstrating particularly strong performance under high noise rates and validating its effectiveness and robustness.

0 citationsRead paper