{"title":"在物理人机交互过程中,从演示、修正和偏好中统一学习","authors":"Shaunak A. Mehta, Dylan P. Losey","doi":"10.1145/3623384","DOIUrl":null,"url":null,"abstract":"Humans can leverage physical interaction to teach robot arms. This physical interaction takes multiple forms depending on the task, the user, and what the robot has learned so far. State-of-the-art approaches focus on learning from a single modality, or combine some interaction types. Some methods do so by assuming that the robot has prior information about the features of the task and the reward structure. By contrast, in this paper we introduce an algorithmic formalism that unites learning from demonstrations, corrections, and preferences. Our approach makes no assumptions about the tasks the human wants to teach the robot; instead, we learn a reward model from scratch by comparing the human’s input to nearby alternatives, i.e., trajectories close to the human’s feedback. We first derive a loss function that trains an ensemble of reward models to match the human’s demonstrations, corrections, and preferences. The type and order of feedback is up to the human teacher: we enable the robot to collect this feedback passively or actively. We then apply constrained optimization to convert our learned reward into a desired robot trajectory. Through simulations and a user study we demonstrate that our proposed approach more accurately learns manipulation tasks from physical human interaction than existing baselines, particularly when the robot is faced with new or unexpected objectives. Videos of our user study are available at: https://youtu.be/FSUJsTYvEKU","PeriodicalId":36515,"journal":{"name":"ACM Transactions on Human-Robot Interaction","volume":null,"pages":null},"PeriodicalIF":4.2000,"publicationDate":"2023-09-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"6","resultStr":"{\"title\":\"Unified Learning from Demonstrations, Corrections, and Preferences during Physical Human-Robot Interaction\",\"authors\":\"Shaunak A. Mehta, Dylan P. Losey\",\"doi\":\"10.1145/3623384\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Humans can leverage physical interaction to teach robot arms. This physical interaction takes multiple forms depending on the task, the user, and what the robot has learned so far. State-of-the-art approaches focus on learning from a single modality, or combine some interaction types. Some methods do so by assuming that the robot has prior information about the features of the task and the reward structure. By contrast, in this paper we introduce an algorithmic formalism that unites learning from demonstrations, corrections, and preferences. Our approach makes no assumptions about the tasks the human wants to teach the robot; instead, we learn a reward model from scratch by comparing the human’s input to nearby alternatives, i.e., trajectories close to the human’s feedback. We first derive a loss function that trains an ensemble of reward models to match the human’s demonstrations, corrections, and preferences. The type and order of feedback is up to the human teacher: we enable the robot to collect this feedback passively or actively. We then apply constrained optimization to convert our learned reward into a desired robot trajectory. Through simulations and a user study we demonstrate that our proposed approach more accurately learns manipulation tasks from physical human interaction than existing baselines, particularly when the robot is faced with new or unexpected objectives. Videos of our user study are available at: https://youtu.be/FSUJsTYvEKU\",\"PeriodicalId\":36515,\"journal\":{\"name\":\"ACM Transactions on Human-Robot Interaction\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":4.2000,\"publicationDate\":\"2023-09-22\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"6\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"ACM Transactions on Human-Robot Interaction\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1145/3623384\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q2\",\"JCRName\":\"ROBOTICS\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"ACM Transactions on Human-Robot Interaction","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1145/3623384","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q2","JCRName":"ROBOTICS","Score":null,"Total":0}
Unified Learning from Demonstrations, Corrections, and Preferences during Physical Human-Robot Interaction
Humans can leverage physical interaction to teach robot arms. This physical interaction takes multiple forms depending on the task, the user, and what the robot has learned so far. State-of-the-art approaches focus on learning from a single modality, or combine some interaction types. Some methods do so by assuming that the robot has prior information about the features of the task and the reward structure. By contrast, in this paper we introduce an algorithmic formalism that unites learning from demonstrations, corrections, and preferences. Our approach makes no assumptions about the tasks the human wants to teach the robot; instead, we learn a reward model from scratch by comparing the human’s input to nearby alternatives, i.e., trajectories close to the human’s feedback. We first derive a loss function that trains an ensemble of reward models to match the human’s demonstrations, corrections, and preferences. The type and order of feedback is up to the human teacher: we enable the robot to collect this feedback passively or actively. We then apply constrained optimization to convert our learned reward into a desired robot trajectory. Through simulations and a user study we demonstrate that our proposed approach more accurately learns manipulation tasks from physical human interaction than existing baselines, particularly when the robot is faced with new or unexpected objectives. Videos of our user study are available at: https://youtu.be/FSUJsTYvEKU
期刊介绍:
ACM Transactions on Human-Robot Interaction (THRI) is a prestigious Gold Open Access journal that aspires to lead the field of human-robot interaction as a top-tier, peer-reviewed, interdisciplinary publication. The journal prioritizes articles that significantly contribute to the current state of the art, enhance overall knowledge, have a broad appeal, and are accessible to a diverse audience. Submissions are expected to meet a high scholarly standard, and authors are encouraged to ensure their research is well-presented, advancing the understanding of human-robot interaction, adding cutting-edge or general insights to the field, or challenging current perspectives in this research domain.
THRI warmly invites well-crafted paper submissions from a variety of disciplines, encompassing robotics, computer science, engineering, design, and the behavioral and social sciences. The scholarly articles published in THRI may cover a range of topics such as the nature of human interactions with robots and robotic technologies, methods to enhance or enable novel forms of interaction, and the societal or organizational impacts of these interactions. The editorial team is also keen on receiving proposals for special issues that focus on specific technical challenges or that apply human-robot interaction research to further areas like social computing, consumer behavior, health, and education.