Qingyang Lyu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
Qingyang Lyu has a total of 6 co-authors. Here are the top co-authors by number of shared publications:
- Jiaqiang Yang
Zhejiang University
2 shared publications - Yasir Salam
College of Electrical Engineering, Zhejiang University, Hangzhou, China
2 shared publications - Rafik Ighil Nessouk
College of Electrical Engineering, Zhejiang University, Hangzhou, China
1 shared publication - Weiang Wang
College of Electrical Engineering, Zhejiang University, Hangzhou, China
1 shared publication Yinbei Li
College of Electrical Engineering, Zhejiang University, Hangzhou, China
1 shared publication- Yinbei Li
College of Electrical Engineering, Zhejiang University, Hangzhou, China
1 shared publication
Latest Publications
- Open Access
- 6 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 190807-190819
Dual-LLM Hierarchical Task Planning and Skill Grounding for Mobile Manipulation in Long-Horizon Restroom Cleaning
Public restroom cleaning is a laborious but critical task for public hygiene, offering both an opportunity and a challenge for robotic automation. Traditional robotic systems struggle in such environments due to partial observability, dynamic conditions, and reliance on rigid task scripts and brittle perception pipelines. This paper presents a hierarchical task planning and skill grounding framework leveraging large language models (LLMs) for autonomous restroom cleaning using a mobile manipulator. A high-level LLM (LLM-H) generates semantic task plans based on symbolic scene representations derived from a visionlanguage model (VLM), while a low-level LLM (LLM-L) refines these plans into a sequence of executable, parameterized robotic skills. To ensure robust and adaptive operation, the framework integrates a semantic active perception module that mitigates visual occlusions through targeted exploration, a dynamic task buffer that reprioritizes tasks in real time to handle environmental changes, and a VLM-based validation module that assesses task outcomes and triggers corrective actions when necessary. Experiments conducted in real-world public restrooms demonstrate that our system outperforms state-of-the-art baselines in task-completion rate while robustly handling partial observability, dynamic events, and long-horizon cleaning sequences. Its modular, scalable approach provides a solid foundation for deploying intelligent service robots in hygiene-critical public spaces.
- 6 November 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 190807-190819
Open AccessDual-LLM Hierarchical Task Planning and Skill Grounding for Mobile Manipulation in Long-Horizon Restroom Cleaning
Public restroom cleaning is a laborious but critical task for public hygiene, offering both an opportunity and a challenge for robotic automation. Traditional robotic systems struggle in such environments due to partial observability, dynamic conditions, and reliance on rigid task scripts and brittle perception pipelines. This paper presents a hierarchical task planning and skill grounding framework leveraging large language models (LLMs) for autonomous restroom cleaning using a mobile manipulator. A high-level LLM (LLM-H) generates semantic task plans based on symbolic scene representations derived from a visionlanguage model (VLM), while a low-level LLM (LLM-L) refines these plans into a sequence of executable, parameterized robotic skills. To ensure robust and adaptive operation, the framework integrates a semantic active perception module that mitigates visual occlusions through targeted exploration, a dynamic task buffer that reprioritizes tasks in real time to handle environmental changes, and a VLM-based validation module that assesses task outcomes and triggers corrective actions when necessary. Experiments conducted in real-world public restrooms demonstrate that our system outperforms state-of-the-art baselines in task-completion rate while robustly handling partial observability, dynamic events, and long-horizon cleaning sequences. Its modular, scalable approach provides a solid foundation for deploying intelligent service robots in hygiene-critical public spaces.
- 6 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (10) , 10266-10273
A Hybrid Framework Using Diffusion Policy and Residual RL for Force-Sensitive Robotic Manipulation
Force-sensitive manipulation is essential for tasks such as cleaning, polishing, and surgical assistance, yet it remains challenging due to complex contact dynamics and the need for real-time adaptation. We propose DP-RRL, a hybrid learning framework that combines a diffusion policy (DP) for imitation learning (IL) with a residual reinforcement learning (RL) module to tackle these challenges. First, the DP leverages expert demonstrations to generate smooth, expert-like motion-force trajectories. Next, a residual RL component refines these trajectories online by adapting force commands to account for unmodeled contact dynamics. Our approach incorporates multimodal perception, including RGB point clouds, force feedback, and proprioceptive data, and achieves robust force-sensitive manipulation in unstructured environments. We validate DP-RRL on a 7-DOF robotic arm performing basin-cleaning tasks, where it demonstrates an average success rate of 88% across 12 unseen scenarios, outperforming state-of-the-art baselines.
- 6 August 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 10 (10) , 10266-10273
A Hybrid Framework Using Diffusion Policy and Residual RL for Force-Sensitive Robotic Manipulation
Force-sensitive manipulation is essential for tasks such as cleaning, polishing, and surgical assistance, yet it remains challenging due to complex contact dynamics and the need for real-time adaptation. We propose DP-RRL, a hybrid learning framework that combines a diffusion policy (DP) for imitation learning (IL) with a residual reinforcement learning (RL) module to tackle these challenges. First, the DP leverages expert demonstrations to generate smooth, expert-like motion-force trajectories. Next, a residual RL component refines these trajectories online by adapting force commands to account for unmodeled contact dynamics. Our approach incorporates multimodal perception, including RGB point clouds, force feedback, and proprioceptive data, and achieves robust force-sensitive manipulation in unstructured environments. We validate DP-RRL on a 7-DOF robotic arm performing basin-cleaning tasks, where it demonstrates an average success rate of 88% across 12 unseen scenarios, outperforming state-of-the-art baselines.