arXiv preprint proposes LLM+RL system for autonomous crane construction using ChatGPT 5
The arXiv preprint (arXiv:2610.26899v1) proposes a method that combines an enhanced soft actor-critic reinforcement learning planner for a mobile crane, robot kinematics/control to execute actions, a custom embedding model to extract environment information, and ChatGPT 5 to provide progress updates, safety information, and reports; the authors report feasibility via several full-scale retaining wall construction experiments. The paper describes algorithmic details and an experimental validation but is a single preprint and its claims are those of the authors.