2017/03/05 by Conor Schenck, Dieter Fox, Schenck, Conor +1
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Innovative Microfluidic and Catalytic Techniques Innovation #Multimodal Machine Learning Applications #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.1703.01564
openalex publication_date 2017/03/05 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Liquids are an important part of many common manipulation tasks in human\nenvironments. If we wish to have robots that can accomplish these types of\ntasks, they must be able to interact with liquids in an intelligent manner. In\nthis paper, we investigate ways for robots to perceive and reason about\nliquids. That is, a robot asks the questions What in the visual data stream is\nliquid? and How can I use that to infer all the potential places where liquid\nmight be? We collected two datasets to evaluate these questions, one using a\nrealistic liquid simulator and another on our robot. We used fully\nconvolutional neural networks to learn to detect and track liquids across\npouring sequences. Our results show that these networks are able to perceive\nand reason about liquids, and that integrating temporal information is\nimportant to performing such tasks well.\n