CRII: CHS: Leveraging Implicit Human Cues to Design Effective Behaviors for Collaborative Robots

通过在制造业、医疗保健和太空探索等关键领域与人类积极合作，机器人有可能显著造福社会。但是，为了提供有效的帮助，机器人必须能够以自然、直觉和善于社交的方式与人合作。当前的人机协作要求人们明确地向机器人伙伴传达他们的目标和期望的响应。因此，人机联合活动与涉及人类团队合作的场景几乎没有相似之处，在这种情况下，人们能够理解伴侣的隐含暗示，如眼神、面部表情和语调，并凭直觉做出适当的反应，如移动到某个位置、先发制人地获取工具或提供澄清。PI在这个项目中的目标是建立一个研究项目，通过开发计算模型来探索协作机器人有效行为的设计，这些计算模型使它们能够感知隐含的人类交流线索，并通过推断线索意图来指导机器人的反应，并评估新算法在人机研究中的有效性。这项研究具有重大的前景，通过帮助实现机器人作为人类工作的关键贡献者、合作伙伴和助手的愿景，使社会受益，应用范围包括家务劳动、制造、建筑、医疗保健和太空探索。除了将项目成果传播给更大的研究社区外，PI还将以他过去成功的外展活动为基础，为K-12暑期课程提供以机器人和计算机科学教育为中心的机会。为此，PI将通过开发一个初步的框架、过程和一套方法来感知和响应隐含的人类交流行为，从而解决设计有效协作机器人的挑战。他的方法将包括(1)观察和分类从事典型协作任务的人类团队的内隐线索和反应，(2)使用从观察到的行为中提取的特征和参数开发目标、线索和反应之间关系的计算模型，(3)整合内隐线索感知和响应算法，以指导特定协作用例中的机器人行为。(4)评估这些行为对协作任务结果的有效性。这项研究将为协作机器人产生一套可推广的设计原则，生成展示实际实现的开源算法，并推进有关人类行为的计算理解的知识。总的来说，这项工作将导致机器人能够更有效地与人合作，并加速辅助机器人融入社会。它将综合人类交流的理论，并探索它们在人机交互中的应用，以及推进关于机器人如何作为人类合作者提供帮助和机器人与人类伙伴密切合作所需的传感器类型的知识。在HRI实验中得到经验验证的隐式感知和响应算法将作为开源机器人操作系统（ROS）的模块传播。

Robots have the potential to significantly benefit society by actively collaborating with people in critical domains including manufacturing, healthcare, and space exploration. But to provide effective assistance, robots must be able to work with people in a natural, intuitive, and socially adept manner. Current human-robot collaborations require that people explicitly communicate their goals and desired responses to robotic partners. As a result, joint human-robot activities bear little resemblance to scenarios involving human-human teamwork, where people are able to understand their partner's implicit cues, such as eye gaze, facial expressions, and intonations, and intuit appropriate responses, such as moving to a certain location, preemptively fetching a tool, or providing a clarification. The PI's goal in this project is to establish a research program that will explore the design of effective behaviors for collaborative robots by developing computational models that enable them to sense implicit human communicative cues and guide robot responses by inferring cue intent, and to evaluate the effectiveness of the new algorithms in human-robot studies. The research holds significant promise of benefiting society by helping to achieve a vision of robots acting as key contributors, partners, and assistants in human work, with applications across a range of activities including domestic housework, manufacturing, construction, healthcare, and space exploration. In addition to disseminating project outcomes to the larger research community, the PI will build on his successful past outreach activities to provide opportunities for K-12 summer programs centered on robotics and computer science education.To these ends, the PI will address the challenge of designing effective collaborative robots by developing a preliminary framework, process, and set of methods to sense and respond to implicit human communicative behaviors. His approach will involve (1) observing and classifying implicit cues and responses for human-human teams engaged in an archetypical collaborative task, (2) developing computational models of the relationships between goals, cues, and responses using features and parameters extracted from observed behaviors, (3) integrating implicit cue sensing and response algorithms to guide robot behaviors in specific collaborative use cases, and (4) evaluating the effectiveness of these behaviors on collaborative task outcomes. This research will produce a set of generalizable design principles for collaborative robots, generate open-source algorithms showcasing practical implementations, and advance knowledge regarding computational understanding of human behaviors. Overall, the work will lead to robots that are able to work more effectively with people and accelerate the integration of assistive robots into society. It will synthesize theories of human communication and explore their application to human-robot interaction, as well as advancing knowledge regarding how robots might provide assistance as human collaborators and the types of sensors necessary for robots working closely with human partners. Implicit sensing and response algorithms that have been empirically validated in HRI experiments will be disseminated as modules for the open-source Robotic Operating System (ROS).