An Integrated Supervised Reinforcement Machine Learning Approach for Automated Clash Resolution

During design coordination, identified relevant clashes are discussed in detail, and design changes and modifications are made to resolve the clashes prior to the construction. Currently, clash resolution is a slow manual process. Recent research focused on using supervised machine learning to automate the clash resolution process shows potential results to improve the efficiency and effectiveness of clash resolution. However, the model trained using supervised learning is limited in its effectiveness by the quality of training data provided. To overcome this limitation, the paper proposes a machine learning method that integrates supervised and reinforcement learning. In the proposed model, supervised learning will be used to establish the initial relationship between the clash information and the clash resolution decision. This relationship will act as pre-training for reinforcement learning, which will improve the relationship with subsequent iterations of the learning process, generating a more effective clash resolution policy than the initial relationship.