Automating the full laundry process is a major challenge in domestic robotics, requiring advances in perception, manipulation, and task coordination. Existing solutions focus on isolated tasks, neglecting transitions and integration into a cohesive pipeline. A key limitation is that robots lack the ability to evaluate their performance and recover from failures, which is essential for real-world deployment. This research develops a general-purpose robot system for fully autonomous laundry handling, addressing two core challenges: (1) efficient learning of subtasks such as sorting, folding, and stacking while ensuring reusability across different garments and scenarios, and (2) robust perception systems that understand garment configurations using semantic information, minimizing the need for re-learning. By addressing these challenges, we can develop a complete robotic laundry pipeline. To tackle these challenges, this work introduces a high-level critic leveraging Vision-Language Models (VLMs) to monitor, evaluate, and guide the robot’s actions. By integrating semantic perception with a high-level critic, this research moves beyond task-specific automation toward a unified, adaptable system. The critic enables robust error handling and learning efficiency, bringing fully autonomous laundry processing closer to reality while advancing robotic manipulation for broader household and industrial applications.