An LLM-enabled Multi-Agent Autonomous Mechatronics Design Framework

Powered by transformer architectures [11, 12] and trained on massive dataset, models such as GPT -4 [13], Claude [14], DeepSeek [15], and PaLM [16] exhibit strong performance in chain-of-thought reasoning [17, 18], few-shot learning [19, 20], and even multimodal understanding when extended to vision-language settings [21, 22]. These capabilities have enabled LLMs to perform not only linguistic tasks, but also to engage in procedural synthesis [23], and structured decision-making [24, 25], laying the foundation for their integration into agent-based systems capable of autonomous planning and tool use. The rise of agent-based systems [26] marks a critical milestone in artificial intelligence, enabling entities to autonomously perceive, reason, and act within specific environments [27]. LLM-driven agents [28] further enhance these capabilities through sophisticated linguistic comprehension and generation, proving effective in diverse applications such as text summarization [29], software debugging [30, 31], documentation automation [32, 33], customer support [34, 35], mathematical theorem synthesis [36, 37], virtual environment navigation [38], and structured data querying [39, 40]. In industry, LLM agents have streamlined narrowly defined workflows, including report generation [41-43] and basic data analytics [44, 45], significantly improving operational efficiency. However, current LLM-based agent implementations remain primarily confined to digital or simulated environments, thus limiting their practical application in complex engineering tasks which require the design of physical embodiment, cross-domain integration, and constraint-aware reasoning. Recent attempts to bridge this gap have emerged, demonstrating initial integration of LLM agents with physical experimentation in domains such as autonomous chemical synthesis [46], materials design [47-49], drug discovery [50], and adaptive multi-agent manufacturing systems [51]. Despite these advancements, little attention has been given to LLM-enabled multi-agent frameworks targeting autonomous mechatronics design, a field inherently requiring multidisciplinary expertise across mechanical engineering, electronics, control systems, and software development.