Self-supervised Transparent Liquid Segmentation for Robotic Pouring

Robots in a position to pour liquids would support in responsibilities like cooking or watering our vegetation. However, clear liquids are tricky to perceive in photographs. A modern paper printed on arXiv.org proposes a technique for perceiving transparent liquid inside of clear containers.

Transparent liquid in a transparent container.

Clear liquid in a transparent container. Image credit: Piqsels, CC0 General public Area

The method employs a generative model that learns to translate visuals of coloured liquid into artificial photographs of a clear liquid, which can be made use of to practice a transparent liquid segmentation product. It does not will need labeled correspondence among coloured visuals and transparent photographs and as a result allows automated and very efficient dataset collection.

Researchers assemble a robotic pouring technique to reveal the utility of the clear liquid segmentation model. Moreover, numerous dataset augmentation experiments are conducted to show the possible of the proposed technique to generalize to varied scenes.

Liquid condition estimation is significant for robotics jobs this kind of as pouring on the other hand, estimating the condition of transparent liquids is a demanding issue. We propose a novel segmentation pipeline that can section clear liquids such as h2o from a static, RGB graphic without the need of requiring any guide annotations or heating of the liquid for training. As an alternative, we use a generative design that is capable of translating visuals of colored liquids into synthetically produced clear liquid visuals, qualified only on an unpaired dataset of colored and clear liquid photos. Segmentation labels of colored liquids are obtained routinely working with background subtraction. Our experiments clearly show that we are able to correctly forecast a segmentation mask for clear liquids with out demanding any handbook annotations. We display the utility of transparent liquid segmentation in a robotic pouring activity that controls pouring by perceiving the liquid top in a transparent cup.

Investigation paper: Narayan Narasimhan, G., Zhang, K., Eisner, B., Lin, X., and Held, D., “Self-supervised Clear Liquid Segmentation for Robotic Pouring”, 2022. Url: https://arxiv.org/ab muscles/2203.01538