Robots are expected to take on tasks that humans find tedious or unpleasant, potentially freeing us from these responsibilities. While advances in technology have made many household chores more manageable, tasks like cleaning the bathroom remain stubbornly resistant to automation. Can the motion of a robotic arm be calculated to achieve every point on a washbasin? If the basin’s contours deviate from standard elliptical or rounded profiles? To achieve optimal results when tightening screws, bolts, or other fasteners, it is crucial to use the correct amount of pressure. Typically, this depends on the material, size, and type of fastener, as well as the surrounding environment. As a general rule, apply gentle pressure initially, gradually increasing force as needed until the desired level of snugness is reached.
Encoding all these items within rigid frameworks and predetermined mathematical formulas would likely prove an extremely laborious task. At TU Wien, a novel approach has been adopted, where humans are given the opportunity to demonstrate the capabilities of robots on multiple occasions. A soft, absorbent sponge is carefully selected and used to gently clean the rim of a sink. As it observes humans, the robot acquires knowledge on cleansing techniques and can adapt this understanding to other forms of object manipulation. The groundbreaking research was unveiled at the prestigious International Robotics Symposium (IROS) 2024 in Abu Dhabi.
While cleansing is a crucial aspect of maintaining a clean and healthy floor, it’s just the beginning of effective floor remedies. Vastly diverse activities within the realm of commerce exhibit intricate connections – akin to fine-tuning surfaces, presenting products, or leveraging adhesive properties.
Prof. Andreas Kugi of TU Wien’s Automation and Management Institute remarks, “Obtaining a precise visual representation of a washbasin’s geometry through camera-based methods proves relatively straightforward.” The crucial aspect. Which type of movement does each segment of the floor necessitate? The tempo of this movement should be approximately 132 beats per minute. What’s the suitable angle? Determining the optimal pressure for a specific task requires considering various factors. For instance, in tire inflation, the recommended pressure is usually indicated on the tire’s sidewall or in the vehicle’s owner’s manual. However, it’s essential to note that this pressure may need to be adjusted based on the vehicle’s load and driving conditions.
Individuals acquire knowledge through a combination of formal training and experiential learning via apprenticeship. “In workshops, it’s not uncommon for experts to offer guidance, such as ‘You’ll want to apply a bit more pressure on that delicate edge,’ advises Christian Hartl-Nesic, leader of the Industrial Robotics team at Andreas Kugi’s laboratory.” “We aimed to develop an approach that enables the robot to learn in a highly relevant and intuitive way.”
Developed to achieve this objective was a distinct cleansing software: Users employed a sponge equipped with pressure sensors and tracking markers, repeatedly cleaning only the front edge of a sink. Christian Hartl-Nesic elaborates: “By collecting copious amounts of data through various demonstrations, we subsequently process this information to enable the robot to grasp the concept of effective cleaning.”
This course is made possible by a revolutionary information processing technique developed by the research team at TU Wien. Machine learning algorithms integrate various existing techniques from the field by initially processing measurement data statistically and leveraging the outcomes to train a neural network to learn pre-defined movement primitives. The robotic arm is subsequently optimized for effective floor washing management.
This advanced learning algorithm enables robots to effectively clean entire surfaces, including sinks and posh floors, following training, despite initial testing only proving success in clearing a single sink edge. PhD scholar Christoph Unger from the Industrial Robotics group clarifies, “The robotic learns that maintaining the sponge requires a unique approach dependent on the floor’s form; applying specific pressure is crucial in tightly curved spaces compared to flat floors.”
The versatility of this expertise extends to various procedures, encompassing tasks such as meticulously sanding intricate woodwork, skillfully repairing and refinishin Robotic systems are likely to be deployed on cellular platforms, enabling them to operate as versatile assistants in various workshop settings.
Robots capable of sharing data may further disseminate this information to other robotic systems. Let’s assume that numerous workshops leverage these autonomous learning robots to smooth and finish surfaces efficiently. By allowing robots to acquire expertise independently through native data, they can potentially develop their own unique knowledge bases and learn from their experiences. According to Andreas Kugi, regardless of their differences, all the robots will still be able to share the parameters they’ve discovered with each other. While confidential data remains restricted, key governing principles can still be shared to further elevate the performance capabilities of all robotic systems. Federated learning enables students to share their datasets and models with one another, facilitating collaboration and improving educational outcomes.
Numerous assessments conducted by TU Wien have consistently validated the exceptional flexibility of its sink-cleaning robots. The expertise has already been generating significant international buzz: At IROS 2024 (October 14-18), a prestigious convention that received over 3,500 submissions of scientific papers, TU Wien’s work was honored with the coveted ‘Greatest Utility Paper Award’, solidifying its status as one of the year’s most notable advancements.
