LLMBot: Multi-Agent Robotic Systems for Adaptive Task Execution

摘要

Task planning for autonomous systems often requires significant computational and financial resources, limiting their accessibility and scalability. In this paper, we have introduced a framework that integrates Large Language Models (LLMs) with autonomous systems for natural language interaction and task execution. The proposed method combines the LLMs for high-level planning with a simulated 3D environment in Unreal Engine and behaviour trees for robust low-level execution of robotic actions. We employ prompt engineering, multi-modal input, and parameter optimization techniques to enhance LLM performance and reduce computational overhead. A comprehensive test suite evaluates success rates, spatial distributions, and cost-effectiveness across various scenarios. The simulation results demonstrate that our approach outperforms traditional planning systems in terms of computational efficiency and cost-effectiveness while maintaining comparable task success rates.