Supervised reinforcement learning to automate mechanical and plumbing trade clash resolution: A preliminary case study

While software like Navisworks has improved the process of conducting clash tests, resolving clashes remains a slow and manual task. Researchers have explored the use of supervised learning to automate clash resolution with successful outcomes. However, the effectiveness of supervised learning in automating tasks is limited by the availability of a large amount of data. To address this limitation, the authors conduct a preliminary case study to understand the feasibility, challenges, and future work required to develop reinforcement learning and supervised-reinforcement learning models towards developing a machine learning model capable of automating clash resolution among mechanical and plumbing trade elements. To achieve the objective of this research, the authors conduct limited training of reinforcement learning and supervised-reinforcement learning models for 1000 episodes towards clash resolution. The limited training showcases a non-monotonous increase in the learning progression of the proposed machine learning algorithms, highlighting the potential feasibility of reinforcement learning towards automating clash resolution. Additionally, the work also highlights how trained supervised learning algorithms can be utilized to initialize the weights for reinforcement learning for a potentially stable learning progression. Based on the limited training conducted and preliminary results of the work undertaken, conclusions are also drawn regarding the feasibility of the proposed machine learning algorithms, the challenges encountered, and directions for future research.