School of Engineering Department of Civil and Environmental Engineering 88 Human-Robot-AI Symbiotic Mobile Mapping Solution for Fast and Regular Examination of Civil and Environmental Engineering Infrastructures Supervisor: WANG Yu-Hsing / CIVL Student: CHAN Kin Yan / CIVL Course: UROP1100, Spring Artificial Intelligence, machine learning, deep learning are trendy techniques widely expending to various fields and providing deeper insight through development. This project aims for exploration of opportunity of the application and implementation of the well-developed machine learning techniques on civil engineering fields, including maintenance of concrete structure and other topics. For exploration and knowledge development purpose, full model development process from data collection, data processing, model training, prediction performing, model fine-tunning and model performance assessment was studied. Then various possible machine learning library including sklearn, tensorflow Keras for both supervised learning and unsupervised learning requirement was explored in detail for opportunity of merging with domain knowledge. U-net architecture used for semantic image classification was also investigated and prepared for future projects including classification of transport types. Human-Robot-AI Symbiotic Mobile Mapping Solution for Fast and Regular Examination of Civil and Environmental Engineering Infrastructures Supervisor: WANG Yu-Hsing / CIVL Student: LI Sau Long / CIVL Course: UROP1100, Spring In Hong Kong, more than 30,000 old buildings and infrastructure demand routine checking and maintenance, some of which require urgent repairs. Manual visual inspection remains to be the standard procedure in the civil engineering and related industries, which could be prone to errors, time-consuming and labour-intensive. Compounded by insufficient manpower, economic losses and safety issues, current practices might prove to be inadequate to meet the city’s operational demands. In this project, a Convolutional Neural Network architecture is utilised to analyse and categorise concrete defects into 3 different levels of damage, in order to speed up the inspection process. A state-of-the-art VGG16 model is incorporated into the code which has approximately 138 million parameters which gives significant flexibility to fine-tune. The results of the accuracy of the model is summarised into a confusion matrix and IoU (Intersection over Union) scores to assess the performance of the model on future projects and identify potential errors sources.