Wei Li

Wei Li (李巍)

I am Wei Li, a third-year Ph.D. student in the School of Computer Science and Technology, Harbin Institute of Technology (Shenzhen), advised by Prof. Rui Shao. My research interests include Computer Vision, Vision-Language Models, and Vision-Language–Action Models.

Email: liwei2024@stu.hit.edu.cn / Google Scholar / Github

Preprints (* equal contribution, † corresponding author)

	World-Calibrated Proposal-to-Action Flow for Vision-Language-Action Models Jie He, Wei Li*, Junwen Tong, Rui Shao, Wei-Shi Zheng, Liqiang Nie arXiv / project page ProAct is a world-calibrated proposal-to-action framework that makes the generative source of flow-based VLA policies predictable via a Proposal Expert and a prospective World Expert.
	UniMPA: A Unified Memory-Prediction-Action Model via Action-Grounded Transition Modeling Wei Li, Rui Shao, Jie He, Lingsen Zhang, Ziwei Liu, Liqiang Nie arXiv / project page UniMPA is a unified memory-prediction-action model that addresses the transition realizability gap by jointly resolving transition ambiguity, prediction–execution mismatch, and experience–realization mismatch through an action-grounded transition interface.
	Learning to Use Imagination: Progress-Conditioned Future Utilization for World Action Models Yijie Zhu, Zitong Yu, Wei Li*, Hui Ma, Wen Li, Rui Shao, Liqiang Nie arXiv / github We propose ProWAM, a Progress-Conditioned World Action Model that introduces execution progress as an explicit intermediate representation for adaptive imagination utilization.
	Large VLM-based Vision-Language-Action Models for Robotic Manipulation: A Survey Rui Shao, Wei Li†, Lingsen Zhang, Renshan Zhang, Zhiyang Liu, Ran Chen, Liqiang Nie arXiv / github This survey provides the first systematic, taxonomy-oriented review of large VLM-based VLA models for robotic manipulation.

Publications (* equal contribution, † corresponding author)

	ATI-VLA: Action-Centric Predictive Vision-Language-Action Models via Actionable Alignment Then Adaptive Injection Yijie Zhu, Rui Shao, Jie He, Wei Li, Bo Zhao, Yelin Wang, Xiaochen Yuan, Tao Tan, Miao Zhang, Xiaojiang Peng, Zitong Yu NeurIPS 2026 arXiv / project page ATI-VLA is an action-centric predictive VLA framework that first aligns predictive observation and action representations in a shared discrete codebook, then adaptively injects the predictive latents into action decoding via a lightweight side-path, achieving state-of-the-art performance with faster convergence.
	ConsisVLA-4D: Advancing Spatiotemporal Consistency in Efficient 3D-Perception and 4D-Reasoning for Robotic Manipulation Wei Li, Jizhihui Liu, Li Yixing, Junwen Tong, Rui Shao, Liqiang Nie CVPR 2026 arXiv / github ConsisVLA-4D optimizes VLA models by ensuring consistency across three dimensions. Semantic: Aligns object identities across multiple viewpoints. Geometric: Resolves spatial ambiguities between objects. Spatiotemporal: Integrates local dynamics and global depth for 4D reasoning.
	SemanticVLA: Semantic-Aligned Sparsification and Enhancement for Efficient Robotic Manipulation Wei Li, Renshan Zhang, Rui Shao, Zhijian Fang, Kaiwen Zhou, Zhuotao Tian, Liqiang Nie AAAI 2026, Oral arXiv / github SemanticVLA formulates vision–language–action learning as a semantic selection–fusion–action coupling process, providing an interpretable and efficient pathway from multimodal perception to robotic action.
	CogVLA: Cognition-Aligned Vision-Language-Action Models via Instruction-Driven Routing & Sparsification Wei Li, Renshan Zhang, Rui Shao, Jie He, Liqiang Nie NeurIPS 2025 arXiv / github We presented CogVLA, a cognition-aligned and instruction-driven Vision-Language-Action framework designed to address the computational inefficiencies and semantic fragmentation in existing VLA models.
	LION-FS: Fast & Slow Video-Language Thinker as Online Video Assistant Wei Li, Bing Hu, Rui Shao, Leyang Shen, Liqiang Nie CVPR 2025 arXiv / github LION-FS adopts a two-stage optimization strategy: 1) Fast Path: Routing-Based Response Determination evaluates frame-by-frame whether an immediate response is necessary. 2) Slow Path: Multi-granularity Keyframe Augmentation optimizes keyframes during response generation.
	SCOOT: Self-supervised Centric Open-set Object Tracking Wei Li, Weiliang Meng, Bowen Li, Jiguang Zhang, Xiaopeng Zhang SIGGRAPH Asia 2023, Poster page SCOOT unifies self-supervised appearance learning, textual-visual feature fusion, and reconstruction-driven association to advance open-set object tracking.