CAREER: Towards Robots that Learn from Everyday Users

与人合作的机器人的发展对于制造业、医药、医疗保健、军事和消费品等多种行业至关重要。实现这一目标的关键是开发能够适应不断变化的任务和用户需求的机器人技术。随着可用性需求的变化——无论是由于制造过程的修改，引入新患者，还是搬迁到新家——技术技能有限的用户必须有能力定制机器人系统的功能。该项目将提供新的理论模型、技术和开源实现，使用户能够在没有编程的情况下有效地与机器人交流高级任务知识。智力优势：该项目将利用人机交互来提高目标识别、特征选择和策略学习算法的效率，并开发数据重用和算法评估的新技术。将通过开发创新的双现实框架来验证假设，用于互联网规模的快速评估和测试。理论将在两个机器人应用领域进行测试：家庭辅助和制造业。该项目的成果将是一个独立于领域的交互式学习框架，能够从与单个非专业用户的简短交互中进行自适应对象识别、特征选择和策略学习。更广泛的影响：这项研究的长期目标是有一天让个人机器人进入日常生活。这项研究将贡献新的理论模型、技术和开源实现，这将加速与人类一起工作的机器人的开发和采用。由此产生的自适应交互系统的理论和方法将对机器人以外的软件系统，如个人电脑和移动设备，产生潜在的社会影响。为了促进这一领域的研究，研究者将把所有开发的软件组件作为开源提供。这项研究还将纳入本科和研究生阶段的课程，以及面向初高中学生的推广活动。基于网络的评估框架将提供一个独特的机会，通过提供尖端机器人技术并使他们能够为其发展做出贡献，来教育公众。

The development of robots that work cooperatively with people is of critical importance for industries as diverse as manufacturing, medicine, healthcare, military, and consumer products. Key to this goal is the development of robotic technologies that are adaptable to changing task and user needs. As usability demands change - whether due to modifications in the manufacturing process, the introduction of a new patient, or the relocation to a new home - users with limited technical skills must have the ability to customize the functionality of robotic systems. This project will contribute new theoretical models, techniques and open source implementations that will enable users to effectively communicate high level task knowledge to robots without programming. Intellectual merit: The project will leverage human-robot interaction to improve the efficiency of object recognition, feature selection and policy learning algorithms, as well as develop new techniques for data reuse and algorithm evaluation. Hypotheses will be validated through the development of an innovative dual-reality framework for Internet-scale rapid evaluation and testing. Theories will be tested in two robotic applications areas: home assistance and manufacturing. The outcome of this project will be a domain independent interactive learning framework capable of adaptive object recognition, feature selection and policy learning from brief interactions with a single, non-expert user. Broader impact: The long term goal of this research is to one day make personal robots accessible to everyday people. This research will contribute new theoretical models, techniques and open source implementations that will accelerate the development and adoption of robots that work alongside people. The resulting theories and methodologies for adaptive interactive systems will have potential societal impact beyond robotics, in software systems such as personal computers and mobile devices. To promote research in this area, the investigator will make all developed software components available as open source. The research will also be integrated into courses at the undergraduate and graduate levels, as well as into outreach activities for middle and high school students. The web-based evaluation framework will provide a unique opportunity to educate the general public by providing access to cutting-edge robotic technology and empowering them to contribute to its development.