Need laundry folded? Don't ask a robot (knowablemagazine.org)

Despite advancements in robotics reminiscent of Rosie the Robot from *The Jetsons*, the simple task of folding laundry remains a significant challenge for AI and machine learning (ML) systems. Researchers have found that while humans intuitively understand how different fabrics respond to manipulation based on real-world experience, robots struggle to apply their training to dynamic, complex scenarios. Traditional robotic approaches often use a “pick and place” method that fails to adapt to unique fabric shapes and weights, resulting in poor performance; a 2024 study highlighted that this method achieved only a 0.41 score on the Intersection over Union (IoU) metric for successful folds. In response, innovative techniques like AdaFold have emerged, which allow robots to adapt their folding strategies in real-time by continuously monitoring the fabric's changes as they manipulate it. This method scored significantly higher, achieving an IoU of 0.83. The ability of AdaFold to adjust to varying conditions shows promise for not just improving laundry folding capabilities but also for enhancing robotic flexibility in diverse tasks. As researchers develop more sophisticated data sets and models, such as ClothesNet with its detailed 3D clothing representations, the dream of efficient robotic household helpers may inch closer to reality, even if the challenges remain formidable.