Visual Imitation Learning for Robotic Manipulation in Cluttered Environments

This thesis explores how robots can learn manipulation tasks in complex environments through visual observation rather than relying on step-by-step programming. The research aims to enable robots to operate effectively in crowded and unpredictable scenarios, such as cluttered workspaces. By learning to interpret visual information and select more efficient actions, the proposed methods enhance robots’ ability to grasp objects and complete tasks while reducing errors. This work contributes to lowering the time and cost required to deploy robotic systems and supports safer and more flexible use of robots in real-world applications.