AI-Empowered Robotics for Manufacturing and Warehouse Item Handling
This research explores the development of AI-powered robotic systems that transcend the traditional role of rigid, pre-programmed machines. It addresses the growing demand for intelligent and adaptive automation in modern industrial environments, which are increasingly defined by high variability and product customization. As manufacturing and logistics shift toward high-mix, low-volume production, conventional automation, reliant on precise models and fixed routines, has become progressively insufficient. In response, this research advances the field of intelligent robotics by integrating artificial intelligence, multi-modal sensing, and human-inspired cognitive capabilities, enabling robots to perceive, reason, and learn within complex and dynamic settings. ; The pursuit of transforming robots into more autonomous agents is examined across three application domains: robotic surface polishing, deburring of deformable shoe soles, and object grasping in warehouse environments. Each scenario demands a high degree of perception, dexterity, and adaptability, qualities typically associated with human skill. ; In manufacturing, a polishing path-planning algorithm is developed for free-form poly-surfaces, allowing robots to perform finishing operations even when 3D models are unavailable. In the footwear industry, where rubber soles vary in shape, color, and material, two deburring setups are investigated: one with a fixed deburring tool and moving workpiece, and another with a fixed workpiece and moving tool. For the first configuration, a hybrid learning-from-demonstration framework is proposed, enabling robots to imitate expert actions through visual segmentation and motion modeling. This is further extended with a self-supervised vision-motion learning method that leverages reinforcement learning to associate visual inputs with tool trajectories, eliminating the need for labeled data or CAD models. For the second setup, a complete processing pipeline is developed, encompassing defect detection and execution, and incorporating online vision-based path correction to compensate for workpiece deformation. ; In the warehouse domain, a grasp learning framework is proposed that enables robots to acquire grasping strategies through experience and adapt them to novel yet similar objects. ; Overall, the proposed methodologies integrate AI, vision, and control to foster flexible and robust robotic behavior, reducing dependence on expert programming while enhancing robotic autonomy.
Linkedin Page
This is my Linkedin Page
Personal Website
This is my Personal Website
Back to Alumni