Machine-Learning Robot Irons Your Clothes
Columbia University researchers used simulations, machine learning and GPUs to teach this robot how to iron
Laundry is easily one of everyone’s least favorite chores. It takes way too long to wash, fold, and iron everything. Well, it appears this tedious task is also going the way of the robots.
We’ve already shown you Laundroid, a laundry-folding robot that uses image recognition and robotic arms to, albeit painstakingly slow, pick and fold a load of laundry. It takes Laundroid 3-10 minutes to fold each item, so it’s best to do this overnight.
But don’t sweat it, if some items folded by Laundroid come out wrinkly, Columbia University researchers have spent the last three years building a machine-learning robot that can iron clothes.
Here’s how it works. The ironing robot picks up a piece of clothing and rotates it 360 degrees to expose it to a Microsoft Xbox Kinect sensor that digitally reconstructs the item.
GPU’s from NVIDIA are used to speed up this process. Once the robot knows what the item is, it knows how to position it for ironing.
The ironing robot fuses two types of surface scans to get the wrinkles out. A curvature scan estimates the height deviation of the cloth surface, while a discontinuity scan can detect sharp surface features, such as wrinkles. The machine-learning robot then uses this information to detect the areas on the clothing that need to be ironed.
Podcast: How NVIDIA’s Jetson TX1 is Making Robots Smarter
An example of robotic ironing with an optimized path. In this case, three permanent wrinkles are detected. (A): Green line segments are detected permanent wrinkles. White circle is the starting position and white arrows show the path orientation. (B): Results after ironing and manually flattening. (C)-(H): Key frames of the whole ironing process in an order.
“Robotic ironing is a very challenging task,” said Yinxiao Li, lead author on a paper set to be published at the IEEE International Conference on Robotics. Li admits that an ironing robot is still too slow and costly to be practical, but he said the robot’s ability to handle floppy, unpredictable objects has lots of applications outside the laundry room. For example, this could help robots do industrial jobs that involve ropes or cable harnesses, or they could be useful in food production.