Too glam to give a(i) damn: the robot stylist
The latest advances in AI have manifested in the form of a robot-assisted dressing system that can successfully dress individuals with different clothes, body shapes and arm shapes.
You’ve probably clicked on this article because the idea of getting dolled up by a robot sounds quite strange, if not totally futuristic. But robot-assisted dressing could greatly improve the well-being and quality of life of those with disabilities or diseases that impact their ability to dress themselves.
We all take our youthful agility and dexterity for granted when we are young. Children particularly often seem to be completely immune to injury, while managing to escape unscathed from impossibly dangerous situations involving stairs, windows or sharp objects.
However, our body can grow less forgiving with age.
Physical decline is one of the most frustrating outcomes of aging, and can be immensely discouraging when even the most simple of daily tasks such as getting dressed and undressed become challenging. According to the National center for Health Statistics, 92% of at-home care patients and nursing facility residents need help with dressing.
Now, a research team at Carnegie Mellon University’s Robotics Institute (RI) (PA, USA) have designed a robot-assisted dressing system that could help relieve the pressure on nursing facilities.
How? They used a stimulation to teach the robot to maneuver clothing and dress people. They then trained it in real world application using reinforcement learning, meaning the robot was positively rewarded for completing specific tasks, such as precisely positioning an article of clothing further along a participant’s arm, leading to gradual success improvement.
The team tested their design using 17 participants and a total of 510 dressing tests (30 per participant), and discovered that the robot successfully dressed 86% of the length of the participants’ arms when averaged across all dressing trials.
This research is remarkable for several reasons. Firstly, previous and existing robot-assisted dressing research has focused on dressing people with one garment option, such as a hospital gown. Secondly, these attempts have also been restricted in their variety of arm positions. The system created by researchers at RI overcomes both these limitations: it was trained to dress participants with different body shapes, arm poses and with five different garments. In fact, the robot even successfully dressed participants while they were constantly moving their arm, albeit in small movements.
But what if the garment falls or twists or moves? And what if the robot is too rough? Well, the team took these considerations into account when creating their system. Because clothes are not stable or fixed objects, the RI researchers used high-dimensional representation for deformable objects to enable the robot to recognize the current state of the clothes and how they interact with a person’s arm – a representation called a segmented point cloud. And to ensure that the robot interacted safely with humans and avoided any excessive force, they applied positive reinforcement for the robot’s gentleness.
The research still requires significant development, but at present the results are extremely exciting. In future, the team aims to improve the system so that the robot is able to accomplish more complicated maneuvers, such as zipping, buttoning, pulling a t-shirt over a person’s head or placing a coat on both arms. The team also hopes to tailor the robot’s abilities to the range of needs required by individuals through carrying out observations in nursing facilities.
Don’t believe it or want to find out more? You can watch videos from the research trials here: https://sites.google.com/view/one-policy-dress/