Qi Lv

Qi Lv (吕奇)

I am a third-year Ph.D. student at Harbin Institute of Technology (Shenzhen) and Great Bay University, advised by Liqiang Nie (IARP Fellow), and Michael Yu Wang (IEEE/ASME/HKIE Fellow), and co-supervised by Xiang Deng. Currently I am a visiting student at the National University of Singapore, working with Mike Shou. Before, I received M.S and B.E from Soochow University.

My research focuses on embodied AI and natural language processing, particularly on integrating multimodal large language models with robotic systems. I am interested in enabling robots to better perceive, reason, and act in the physical world through language-vision-action alignment.

Email / CV / Scholar / Twitter / Github / LinkedIn

Selected Publication (* denotes equal contribution)

	InternVLA-A1: Unifying Understanding, Generation and Action for Robotic Manipulation InternVLA-A1 Team arxiv, 2025 project page / arXiv / code / data InternVLA-A1 unifies scene understanding, visual foresight, and action execution into a single framework. Moreover, it is empowered by high-fidelity synthetic data (InternData-A1).
	F1: A Vision-Language-Action Model Bridging Understanding and Generation to Actions Qi Lv, Weijie Kong, Hao Li, Jia Zeng, Zherui Qiu, Delin Qu, Haoming Song, Qizhi Chen, Xiang Deng, Jiangmiao Pang arxiv*, 2025 project page / arXiv / code A novel paradigm integrating visual foresight generation into the decision-making pipeline, enabling robots to plan and execute complex tasks in dynamic environments.
	Cortical Policy: A Dual-Stream View Transformer for Robotic Manipulation Xuening Zhang, Xiang Deng, Qi Lv, Miao Zhang, Xingbo Liu, Liqiang Nie ICLR, 2026 Openreview We propose Cortical Policy, a cortex-inspired dual-stream view transformer for robotic manipulation that jointly reasons over static views (for 3D spatial understanding via pretrained 3D keypoint alignment) and dynamic views (for adaptive behavior via egocentric gaze–aware pretraining).
	STAR: Learning Diverse Robot Skill Abstractions through Rotation-Augmented Vector Quantization Hao Li, Qi Lv, Rui Shao, Xiang Deng, Yinchuan Li, Jianye Hao, Liqiang Nie ICML, 2025, Spotlight arXiv / code A framework for robust skill learning and composition that mitigates codebook collapse and models causal dependencies between skills, by introducing rotation-augmented skill quantization and a causal skill transformer.
	Spatial-Temporal Graph Diffusion Policy with Kinematic Modeling for Bimanual Robotic Manipulation Qi Lv, Hao Li, Xiang Deng, Rui Shao, Yinchuan Li, Jianye Hao, Longxiang Gao, Michael Yu Wang, Liqiang Nie CVPR, 2025 arXiv KStar Diffuser leverages dynamic spatial-temporal robot graphs and differentiable kinematics to produce physically feasible, structure-aware bimanual manipulation policies.
	3D-AffordanceLLM: Harnessing Large Language Models for Open-Vocabulary Affordance Detection in 3D Worlds Hengshuo Chu, Xiang Deng, Qi Lv, Xiaoyang Chen, Yinchuan Li, Jianye Hao, Liqiang Nie ICLR, 2025 arXiv We formulates 3D affordance detection as a language-driven reasoning task for open-world scenes. By integrating LLM reasoning and multi-stage training, it achieves improved open-vocabulary affordance segmentation.
	Decision Mamba: A Multi-Grained State Space Model with Self-Evolution Regularization for Offline RL Qi Lv, Xiang Deng, Gongwei Chen, Michael Yu Wang, Liqiang Nie NeurIPS, 2024 arXiv / code We introduces a novel multi-grained Mamba architecture that jointly models historical hidden states and intra-step RTG–state–action relations, enabling more robust offline RL under out-of-distribution settings.
	RoboMP²: A Robotic Multimodal Perception-Planning Framework with Multimodal Large Language Models Qi Lv, Hao Li, Xiang Deng, Rui Shao, Michael Yu Wang, Liqiang Nie ICML, 2024 project page / arXiv / code A multimodal perception–planning framework that grounds MLLM reasoning in embodied robotic manipulation. It improves generalization by combining goal-conditioned perception with retrieval-augmented planning.