The rapid increase of the world population has made traditional construction methods insufficient to meet housing demand. Advances in machine learning have encouraged researchers to apply these techniques to the construction sector, enabling smarter and more efficient building processes. In recent years, the emergence of large language models (LLMs) has opened new possibilities for transforming conventional construction into intelligent construction. This paper proposes an autonomous construction method that integrates a large language model with reinforcement learning.
An enhanced Soft Actor‑Critic (SAC) reinforcement learning algorithm is developed to plan actions for a mobile crane based on construction tasks. Robot kinematics and control algorithms ensure that the actions generated by the RL agent can be executed accurately by the crane.
A customized embedding model is built to extract relevant information from the construction environment according to user interests. Then, the large language model ChatGPT 5 is employed to update construction progress, provide safety information to the control system, and automatically generate progress reports.
The feasibility of the proposed approach is validated through several full‑scale retaining‑wall construction experiments, demonstrating autonomous planning, real‑time safety monitoring, and high‑quality report generation.
Review: This work successfully combines the language understanding capabilities of LLMs with the decision‑making strengths of reinforcement learning, offering a practical pathway for construction automation, especially in complex site environments where safety and progress management are critical.