基于大规模语言模型的建筑组装机器人闭环控制方法
By fine-tuning a large-scale language model in layers and building a custom API library, closed-loop control of a construction assembly robot was achieved, solving the open-loop problem of control programs in existing methods, improving the accuracy and efficiency of code generation, and simplifying the programming process.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- HUAZHONG UNIV OF SCI & TECH
- Filing Date
- 2024-01-05
- Publication Date
- 2026-07-17
AI Technical Summary
Existing robot control methods based on large-scale language models have open-loop characteristics, making it impossible to achieve global optimization of the control program. Furthermore, existing methods rely on predefined robot control skill interfaces, which may lead to errors and safety issues under different robot configurations and building assembly tasks.
A hierarchical strategy is adopted to fine-tune the large-scale language model for instruction and low-rank adaptation, and a custom robot API library is built. By describing and interpreting control tasks at different levels through natural language, closed-loop control programs are generated, and they are evaluated and iteratively optimized in robot simulation platforms and real environments.
It realizes the automatic generation of control code for building assembly robots and human-oriented closed-loop control, which improves the accuracy and efficiency of code generation, simplifies programming tasks, and supports rapid adaptation and global optimization for different assembly tasks.
Smart Images
Figure CN118024237B_ABST