CHS: Small: Formal Design of Human Robot Collaboration in Safety Critical Scenarios

人机协作技术旨在将人类的优势与机器人的优势结合起来。机器人擅长以更高的精度和速度处理重复的程序，并且耐力更长。另一方面，人类拥有优越的感知能力，在面对不确定性和意外情况时表现得更好。例如，透明玻璃上的划痕可以很容易地被人眼检测到，但对计算机视觉来说却是一个极其困难的挑战。寻找有助于设计人与机器人（或一般的人与计算机）之间有效协作的原则是网络-人类系统进步的核心。此外，许多网络-人类系统应用，如关节装配制造、驾驶辅助和机器人辅助手术，都是安全关键，需要在保证性能的情况下完成复杂的高级任务。本课题旨在推导出一种可证明正确的人机协作设计理论，保证高水平复杂任务的完成。该项目的研究可以通过增加许多涉及人机和人机协作的实际应用中的安全性和可信度来造福社会，例如服务机器人，自动化制造系统，应急响应和未知空间的探索。本项目采用向量自回归部分可观察马尔可夫决策过程（VAR-POMDP）模型来管理不确定性，并使用非参数贝叶斯方法从数据中学习模型。利用学习到的模型，研究了基于形式规范的人机协作高级任务自动规划问题。研究小组将进一步研究如何在机器人与不同个体互动或面临不确定环境时实现整个系统的在线（实时）适应。除了理论研究，该团队还将开发软件工具，并通过真实的机器人试验台评估设计理论的有效性。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Human-robot collaboration technologies aim to combine the strengths from humans with those of robots. Robots excel at handling repeated routines with much higher precision and speed, and longer endurance. Humans, on the other hand, have superior perception capabilities and are much better in face of uncertainties and unexpected situations. For example, a scratch on a transparent glass can be easily detected by human eyes but presents an extremely hard challenge for computer vision. Finding principles to help design effective collaboration between humans and robots (or humans and computers in general) is core to advances in cyber-human systems. In addition, many cyber-human system applications, such as joint assembly manufacturing, driver assistance, and robot-assisted surgery are safety critical and need to achieve complex high-level tasks with guaranteed performance. This project aims to derive a provably-correct human-robot collaboration design theory that can guarantee the accomplishment of high-level complex missions. Research from this project can benefit society by increasing the safety and trustworthiness in the many real-world applications involving human-robot and human-computer collaborations such as service robots, automated manufacturing systems, emergency responses, and exploration of unknown spaces.This project adopts a new model, called vector auto-regressive partially observable Markov decision process (VAR-POMDP), to manage uncertainties, and it uses non-parametric Bayesian methods to learn the model from data. With the learned model, an automatic high-level task planning in human-robot collaboration with respect to formal specifications is studied. The team of researchers will further study how to achieve online (real-time) adaptations of the overall system when robots are interacting with different individuals or facing uncertain environments. Beyond theoretical studies, the team will develop software tools and evaluate the effectiveness of the design theory through a real robotic test-bed.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.