Sudharshan Suresh, Haozhi Qi, Tingfan Wu, Taosha Fan, Luis Pineda, Mike Lambeta, Jitendra Malik, Mrinal Kalakrishnan, Roberto Calandra, Michael Kaess, Joseph Ortiz, Mustafa Mukadam
{"title":"NeuralFeels with neural fields: Visuotactile perception for in-hand manipulation","authors":"Sudharshan Suresh, Haozhi Qi, Tingfan Wu, Taosha Fan, Luis Pineda, Mike Lambeta, Jitendra Malik, Mrinal Kalakrishnan, Roberto Calandra, Michael Kaess, Joseph Ortiz, Mustafa Mukadam","doi":"10.1126/scirobotics.adl0628","DOIUrl":null,"url":null,"abstract":"<div >To achieve human-level dexterity, robots must infer spatial awareness from multimodal sensing to reason over contact interactions. During in-hand manipulation of novel objects, such spatial awareness involves estimating the object’s pose and shape. The status quo for in-hand perception primarily uses vision and is restricted to tracking a priori known objects. Moreover, visual occlusion of objects in hand is imminent during manipulation, preventing current systems from pushing beyond tasks without occlusion. We combined vision and touch sensing on a multifingered hand to estimate an object’s pose and shape during in-hand manipulation. Our method, NeuralFeels, encodes object geometry by learning a neural field online and jointly tracks it by optimizing a pose graph problem. We studied multimodal in-hand perception in simulation and the real world, interacting with different objects via a proprioception-driven policy. Our experiments showed final reconstruction <i>F</i> scores of 81% and average pose drifts of 4.7 millimeters, which was further reduced to 2.3 millimeters with known object models. In addition, we observed that, under heavy visual occlusion, we could achieve improvements in tracking up to 94% compared with vision-only methods. Our results demonstrate that touch, at the very least, refines and, at the very best, disambiguates visual estimates during in-hand manipulation. We release our evaluation dataset of 70 experiments, FeelSight, as a step toward benchmarking in this domain. Our neural representation driven by multimodal sensing can serve as a perception backbone toward advancing robot dexterity.</div>","PeriodicalId":56029,"journal":{"name":"Science Robotics","volume":"9 96","pages":""},"PeriodicalIF":26.1000,"publicationDate":"2024-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://www.science.org/doi/reader/10.1126/scirobotics.adl0628","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Science Robotics","FirstCategoryId":"94","ListUrlMain":"https://www.science.org/doi/10.1126/scirobotics.adl0628","RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ROBOTICS","Score":null,"Total":0}
引用次数: 0
Abstract
To achieve human-level dexterity, robots must infer spatial awareness from multimodal sensing to reason over contact interactions. During in-hand manipulation of novel objects, such spatial awareness involves estimating the object’s pose and shape. The status quo for in-hand perception primarily uses vision and is restricted to tracking a priori known objects. Moreover, visual occlusion of objects in hand is imminent during manipulation, preventing current systems from pushing beyond tasks without occlusion. We combined vision and touch sensing on a multifingered hand to estimate an object’s pose and shape during in-hand manipulation. Our method, NeuralFeels, encodes object geometry by learning a neural field online and jointly tracks it by optimizing a pose graph problem. We studied multimodal in-hand perception in simulation and the real world, interacting with different objects via a proprioception-driven policy. Our experiments showed final reconstruction F scores of 81% and average pose drifts of 4.7 millimeters, which was further reduced to 2.3 millimeters with known object models. In addition, we observed that, under heavy visual occlusion, we could achieve improvements in tracking up to 94% compared with vision-only methods. Our results demonstrate that touch, at the very least, refines and, at the very best, disambiguates visual estimates during in-hand manipulation. We release our evaluation dataset of 70 experiments, FeelSight, as a step toward benchmarking in this domain. Our neural representation driven by multimodal sensing can serve as a perception backbone toward advancing robot dexterity.
期刊介绍:
Science Robotics publishes original, peer-reviewed, science- or engineering-based research articles that advance the field of robotics. The journal also features editor-commissioned Reviews. An international team of academic editors holds Science Robotics articles to the same high-quality standard that is the hallmark of the Science family of journals.
Sub-topics include: actuators, advanced materials, artificial Intelligence, autonomous vehicles, bio-inspired design, exoskeletons, fabrication, field robotics, human-robot interaction, humanoids, industrial robotics, kinematics, machine learning, material science, medical technology, motion planning and control, micro- and nano-robotics, multi-robot control, sensors, service robotics, social and ethical issues, soft robotics, and space, planetary and undersea exploration.