The team used image translation to help the robot identify and pour water into a glass. (Image Credit: Carnegie Mellon University)
Robots have difficulty handing water due to its clear quality, making it hard to see. Sure, they poured it in the past, but those techniques involved heating the water and using a thermal camera, which wasn't an efficient solution for everyday applications. Carnegie Mellon University researchers managed to train a robot to identify water, pouring it into a glass thanks to the help of a zebra, horse, image translation, and AI. This could lead to robot servers refilling water glasses, robot gardeners watering plants, or robot pharmacists measuring and mixing medicine.
Image Translations train AI via groups of images, changing an image's style. For example, it can convert a photo into a Monet-style painting or make an image of a horse look like a zebra. The team utilized contrastive learning for unpaired image-to-image translation (CUT) technique.
"You need some way of telling the algorithm what the right and wrong answers are during the training phase of learning," said David Held, an assistant professor in the Robotics Institute who advised Narasimhan. "However, labeling data can be a time-consuming process, especially for teaching a robot to pour water, for which the human might need to label individual water droplets in an image."
"Just like we can train a model to translate an image of a horse to look like a zebra, we can similarly train a model to translate an image of colored liquid into an image of transparent liquid," Held said. "We used this model to enable the robot to understand transparent liquids."
Robots cannot see water very well due to how it reflects, refracts, and absorbs light on different backgrounds. By playing YouTube videos behind a glass filled with water, the team trained the system to see various backgrounds through a glass of water. This training technique enables the robot to pour water against different backgrounds, despite the robot's location.
"Even for humans, sometimes it's hard to precisely identify the boundary between water and air," Narasimhan said.
During experiments, the robot poured water into a glass before it reached a specific point. This process was repeated using glasses of various shapes and sizes. The team also says additional research could be applied to this technique, which includes implementing different lighting conditions, making the robot pour water from one container to another, and estimating the water's height and volume.
Have a story tip? Message me at: http://twitter.com/Cabe_Atwell