Key points are not available for this paper at this time.
We propose and implement a task planning system that integrates a series of validated robotic instructions, deep learning models, and demonstration cases, into a larger structure using the Work Breakdown Structure (WBS) which is commonly used in engineering project management. By organizing these small-scale solutions in a sequence based on temporal and resource dependencies, we use simulations and evaluation functions to select the optimal structure as the Standard Operating Procedure (SOP). These SOPs are then used to train a generative AI, enabling it to mimic human experts in generating high-level structures according to real-world needs and perform minor parameter generalizations to address issues involving temporal dependencies, spatial relationships, and resource allocation. The generated high-level structures are gradually transformed into low-level operations to execute actual tasks. The system records the execution results for automatic or manual analysis, and the feedback is used to continuously improve the generative AI's output. Our work lays the foundation for further research on the innovative application of combining generative AI with WBS data in robotic task planning. https://github.com/NOMIzy/LLM-Project
Yue et al. (Tue,) studied this question.