CHS: Small: Socially-Aware Navigation

机器人以护理和合作的辅助能力提供服务的承诺是无价的。社交辅助机器人有可能对需要物理治疗的人、需要教育干预的儿童、患有发育障碍的儿童和中风后的患者产生积极影响。然而，社交不当行为可能会损害这些辅助平台的长期采用。特别是，虽然通常只被理解为从A点到B点，但机器人的移动可以构成不使用任何语言的通信，甚至当不是有意的时候也是如此。这可能会对一个或多个人与机器人之间的社交互动的成功产生严重的影响。在长期交互中(例如在医院、学校或家庭环境中)，机器人每天可能有数十个甚至数百个导航目的地；有时，“最佳”导航行为可能与社会适当的行动方案不匹配。社会上不适当的导航行为被认为是长期在医院环境中采用辅助机器人的主要障碍。为了解决这一问题，该项目将为自主机器人开发一种社会性感知的导航规划器，它模仿人与人交互的社会惯例，从而推动自主机器人技术的应用。基于模型的社会适宜性导航规划将包括观察社会上接近机器人和/或其他人的人的活动和意图，然后选择空间和社会上都最优的机器人轨迹。然后，这项工作将在此基础上，通过扩大机器人可以理解并以自主方式进行通信的空间通信库来进行。这项研究将把意图识别等高层特征与人际距离等低层特征结合到模型开发中，以创建既考虑空间场景又考虑该场景中演员推断的意图的复杂行为模型。将利用帕累托凹陷消除变换(PaCcET)多目标优化框架来制定适合社会的导航行为政策。开发的规划器将在几种类型的机器人平台上进行验证。

The promise of robots to serve in assistive capacities for care and cooperation is invaluable. Socially assistive robots have the potential to make a positive impact on people in need of physical therapy, children in need of educational intervention, children with developmental disorders, and patients post-stroke. However, socially inappropriate behavior can harm the long-term adoption of these assistive platforms. In particular, while typically understood only as getting from point A to point B, robot movement can constitute communication without any use of words, and even when not intended. This can have serious ramifications for the success of a social interaction between one or more people and a robot. A robot in a long-term interaction (say in a hospital, school, or home setting) may have dozens or even hundreds of navigation destinations every day; sometimes, the "optimal" navigation behavior may not match the socially appropriate course of action. Socially inappropriate navigation behavior has been attributed as a major impediment to the long-term adoption of assistive robots in hospital settings. To address that problem, this project will develop a socially-aware navigation planner for autonomous robots that mimics the social conventions of human-human interactions, and will thereby advance the applications of autonomous robot technology.Model-based navigation planning for social appropriateness will involve observing the activities and intentions of people in social proximity to a robot and/or other people, and then selecting robot trajectories that are both spatially and socially optimal. This work will then build upon that foundation by expanding the library of spatial communication that a robot can understand and communicate in an autonomous fashion. The research will incorporate high-level features, such as intent recognition, into the model development along with low-level features, such as interpersonal distance, to create models of complex behavior considering both the spatial scene and the inferred intention of the actors in that scene. The Pareto Concavity Elimination Transformation (PaCcET) multi-objective optimization framework will be utilized to develop socially appropriate navigation behavior policies. The developed planner will be validated on several types of robot platforms.