Recycling of Waste Electrical and Electronic Equipment (WEEE) is a process still heavily based on manual tasks performed by human workers. HR-Recycler is developing a hybrid human-robot setup to replace currently manual tasks that are largely hazardous, time consuming, and expensive. Through the project, researchers from the Chair of Automatic Control Engineering at the Technical University of Munich have been developing methods that allow the robot to safety and robustly disassemble WEEE into recyclable components. One of the unique approaches to their work is how a robot is used not only for disassembling objects, but also to feel and understand the object characteristics, like a human does.
One of the challenges in robot-based object disassembly is removal of an outer shell or casing – these are common in products such as PC towers, microwave ovens and emergency lamps. The outer shells of modern devices are usually held together with multiple fixtures (e.g., screws). The precise locations and types of fixtures vary greatly between objects. When disassembling devices without knowing their construction details, fixtures are easily missed – either because they are not visible, such as snap-in fixtures, or because they are not properly recognized, for example when screws are recessed. Despite the impressive progress in computer vision techniques, visual data can only provide very limited information about materials and their properties, inside structures and general mechanical design.
For unknown object exploration, humans do not solely rely on visual information, but also take further sensory information into account. It is promising to analyze and reproduce human behavior. Mimicking the human way of exploring unknown objects from haptic information bears a high potential towards improving robotic perception for unknown and complex objects. Using haptic feedback as an alternative source of information further allows improving the knowledge about the surroundings of the robot in terms of actually feeling the properties of objects. This approach, known as “haptic exploration”, significantly improves object identification over purely vision-based methods.
Thus, the challenge of exploring and identifying unknown objects is tackled by augmenting uncertain prior knowledge from the vision data with haptic information. The location of missing fixtures is determined through probing the case structure by bending it from multiple locations in order to identify a model for the structure and hidden fixtures like a human would do.