Published by: Institute of Robotics and Intelligent Systems Editor: Chen Machuan
WUST News - Recently, the Collaborative Robotics Team at the Institute of Robotics and Intelligent Systems of the Wuhan University of Science and Technology (WUST) has achieved significant breakthroughs in the areas of large models and embodied intelligence research. Two of their research papers have been accepted for presentation at the prestigious International Conference on Intelligent Robots and Systems (IROS 2025).
Led by Professor Min Huasong, the team has made substantial progress in two key areas: autonomous robot operation leveraging large language models, and myoelectric control of prosthetic hands enhanced with embodied intelligence.
Autonomous Robot Operations with Large Language Models
In the realm of autonomous robot operation, Associate Professor Lin Yunhan, Ph.D. student Zhou Haotian, and their colleagues have focused on the integration of multimodal perception, large model reasoning, and cognitive decision-making. They pioneered a novel "Robot Reasoning and Autonomous Operation Framework Based on Large Language Models and Behavior Trees (LLM-BT)". This framework integrates visual, linguistic, and action modalities to handle external disturbances and autonomously execute tasks in complex environments, significantly enhancing the decision-making capabilities of robots. To address the limitations of traditional methods in learning capacity and the manual construction of action libraries, the team has innovatively introduced the "Human-in-the-Loop Learning" mechanism. This mechanism generates action knowledge through large language models, refines it through multiple rounds of human-machine interaction, and ultimately converts it into an expandable behavior tree, effectively improving system learning performance, enabling robots to complete tasks despite external interferences. This line of research was previously published at the top robotics conference ICRA in 2024 and further elaborated in the flagship journal Robotics and Autonomous Systems in 2025. The latest results have been accepted by IROS 2025.
Robotic Grasping with Natural Language and Commonsense Reasoning
Another notable achievement from the team, completed by Lin Yunhan and master’s student Wu Wenqi, introduced the "LangGrasp" framework for robotic grasping based on natural language interaction and large language model commonsense reasoning. This framework refines large language models to extract implicit intentions from language instructions, combining them with a point cloud localization module to achieve precise part-level localization within scenes. By extending grasping operations from object-level to part-level, LangGrasp significantly enhances the quality and accuracy of six-degree-of-freedom grasping postures for target objects. The related paper has also been accepted by IROS 2025.
Embodied Intelligence for Myoelectric Prosthetic Hands
In the field of embodied intelligence, Professor Min Huasong's team has collaborated extensively with Professor Sun Fuchun from Tsinghua University and Professor Wen Li from Beihang University. Together, they have developed a novel soft myoelectric prosthetic hand designed to address the operational difficulties and quality- of-life issues faced by amputees when using computer peripherals such as mice and keyboards. This prosthetic hand, featuring dexterous soft robotic fingers and a "one-to-many" mapping myoelectric interaction interface, effectively improves the efficiency of amputees in operating physical mice, reduces task completion time, lowers muscle usage, and alleviates control burdens.
The research provide important theoretical and engineering support for enhancing the computer interaction capabilities of amputees and showcase the broad application prospects of soft robotic prosthetic technology.
These findings, with WUST as a co-first author institution, have been published in Advanced Intelligent Systems (cover paper), Soft Robotics, and Tsinghua Science and Technology.
Looking Ahead
In recent years, the team has published over ten papers in leading international journals, robotics conferences, and core SCI journals. Their research achievements have garnered widespread attention, highlighting the university’s strength in both developing cutting-edge intelligent robotics technology and conducting theoretical research. Moving forward, the team will continue to deepen the integration of large models and embodied intelligence research, promoting the application of intelligent robotics technology in a wider range of real-world scenarios. (Institute of Robotics and Intelligent Systems)
