Mutual intention recognition for human-robot cooperation

在传统的自动化系统中，机器人被用于长时间重复执行相同的任务。围栏通常通过防止人类工人进入机器人工作空间来确保他们的安全。最近的进展使安全的人机共存没有这些物理障碍。由于人类和机器人能够在共享工作空间内安全移动，因此可以将人类代理集成到生产过程中。人类和机器人可以利用各自的技能实现共同的目标。有了智能机器人和人类独有的认知技能的共生组合，可以设计出混合和灵活的协作系统，这些系统对许多应用都很有用，例如小批量生产、服务、车间、实验室和家庭。该研究项目的目标是在人机合作概念的背景下开发和评估相互意图识别。在这种情况下，人类和机器人将作为平等的伙伴，可以自由和动态地选择他们的行动。该项目假设合作伙伴可以用非语言的方式传达他们的意图。理想情况下，合作伙伴可以在执行过程中动态协调。一个新开发的任务表示，可以编码的替代方法和观察的不确定性，将形成代表不同的计划的基础。沟通又分为传达讯息与理解讯息，对于机器人而言，意图辨识与易读性的概念可以达到这个目的。隐马尔可夫模型，运动预测和人类感知的运动规划被用作方法。需要回答的核心问题是这些概念如何预测当前的行动和相当长期的计划，以及团队的流畅性如何从概念的相互作用中受益。

In traditional automation systems, robots are used to repeatedly perform the same task for a long period of time. Fences often ensure the safety of human workers by preventing them from entering the robot workspace. Recent advances enable safe human-robot coexistence without these physical barriers. With humans and robots being able to move safely within the shared workspace, it is possible to integrate human agents into the production process. Humans and robots can work on a common goal using their individual skills. With this symbiotic combination of intelligent robotics and the cognitive skills that are unique to humans, hybrid and flexible cooperating systems can be designed that are useful for many applications, such as small-series production, service, workshops, laboratories and households. The goal of the research project is the development and evaluation of mutual intention recognition in the context of a concept for human-robot cooperation. In this context, humans and robots are to act as equal partners that can freely and dynamically choose their actions. The project assumes that the partners can communicate their intentions non-verbally. Ideally, the partners can coordinate dynamically during execution. A newly developed task representation that can encode alternative approaches and observation uncertainties shall form the foundation for representing different plans. Communication is subdivided into communicating a message and understanding it. For the robot, the concepts of intention recognition and legibility shall achieve the goal. Hidden Markov models, motion prediction and human-aware motion planning are used as approaches. Central questions to be answered are how these concepts can predict the current action and fairly long-term plans and how team fluency benefits from the interaction of the concepts.