Researchers at TU Wien (Vienna University of Technology) have developed a robotic arm capable of learning to clean washbasins through imitation. By observing human actions, the robot adapts to various basin shapes and applies appropriate force for effective cleaning. The development addresses challenges in automating tasks like bathroom cleaning, where fixed programming would be inefficient due to the variations in basin design and required pressure at different points.
The team used a specially designed sponge equipped with force sensors and tracking markers, allowing a human to demonstrate the cleaning of a basin’s edge. This process generated extensive data, which was processed through machine learning techniques to identify essential movement patterns, known as “motion primitives.” These learned actions enable the robot to clean entire sinks and potentially other complex surfaces. Published at IROS 2024, an international robotics conference in Abu Dhabi, the research earned the “Best Application Paper Award” for its application potential in industrial surface treatment.
Beyond cleaning, this technology has implications for tasks such as sanding, polishing, and painting in industrial settings. According to Andreas Kugi, a professor at TU Wien’s Automation and Control Institute, this robotic system could allow robots to learn similarly to human apprentices, gaining skills through demonstrated movements. The research team, led by Kugi and Christian Hartl-Nesic, developed a learning model that integrates several machine learning techniques to translate observational data into robotic action, allowing robots to adapt their cleaning approach based on surface geometry and force requirements.
Future applications may see robots on mobile platforms performing various tasks in workshops and sharing learning data with other robots through a method called “federated learning.” This would allow individual robots to retain data specific to local objects while contributing to a collective knowledge base, enhancing their capabilities across multiple settings.
