Background
Breakthroughs in natural language understanding by large language models (LLMs) open new possibilities for industrial automation. This project uses LLMs as the core for semantic command parsing, driving automated field task assignment and validation to ensure standardized execution at each human-robot collaboration node.
Research Objectives
Through semantic understanding technology, operator natural language commands are translated into structured task sequences executable by robots, while establishing a real-time task execution verification feedback mechanism.
- Develop an LLM-based semantic command parsing engine
- Build an automated task assignment and priority scheduling system
- Design an execution status monitoring and anomaly detection module
Technical Approach
The research applies Instruction Fine-tuning and Retrieval-Augmented Generation (RAG) to enable LLMs to accurately understand domain-specific operational command semantics and map them to robot OS action primitives.
Expected Outcomes
A LLM semantic command-driven middleware integrable with the existing ROS2 ecosystem will be completed, with 500 command execution accuracy validations at the test site targeting over 95% success rate.