Supervised By: Dr. Zhongguo Li, Lecturer (Teaching) in Robotics & AI (UCL EAST), Dept of Computer Science, Faculty of Engineering Science, UCL.
Prof Simon Julier, Professor of Situation Awareness Systems, Dept of Computer Science, Faculty of Engineering Science, UCL.
Locating and estimating the source of a chemical release is a critical task in various fields, including environmental monitoring, industrial safety, and emergency response. However, traditional methods are often limited by their inability to adapt to complex and dynamic environments. This project presents a novel approach for Chemical Source Estimation and Localization using Active Deep Reinforcement Learning. The proposed methodology aims to address the problem of chemical source estimation and localization by modelling a Gaussian plume and adding noise to it to simulate real-world conditions. A particle filter is applied to estimate the source location based on noisy sensor measurements. Subsequently, Active Deep Reinforcement learning is employed to find the most optimal path from the initial position of the robot to the estimated source location. In Active Reinforcement Learning, the agent must determine its actions since there is no predetermined policy for it to follow. The approach utilizes an active learning strategy to analyse a trade-off between exploration and exploitation. This research offers a promising approach for autonomous chemical sensing and source tracking and can have applications in search and rescue operations and disaster management.