NRI: FND: Safe and Efficient Robot Collaboration System (SERoCS) for Next Generation Intelligent Industrial Co-Robots

在目前的自动化工厂中，人类和机器人通常是分开工作的，部分是出于安全原因，部分是因为完全机器人自动化一直是一个目标。然而，近年来，人们已经认识到，当机器人被带出笼子，并允许在共享工作空间与人类工人合作时，存在着巨大的机会。这种协作既利用了人类的智慧、适应性和灵活性，也利用了机器人的耐力、力量和可靠性。在人类和机器人之间的任何协作中，重要的是考虑和确保人类的安全和机器人的最佳性能。该项目旨在为安全高效的机器人协作系统(SERoCS)建立一套设计原则。在工厂之外，SERoCS还可以应用于其他环境中，例如在机器人帮助人类移动的情况下，以及人类驾驶车辆和自动驾驶车辆共享一条道路的自动驾驶场景中。SERoCS由三部分组成：(1)用于环境监测的稳健认知算法，(2)用于安全人-机器人协作的最优任务规划算法，(3)用于安全人-机器人交互(HRI)的安全运动规划和控制算法。认知环境监测算法的研究包括认知模型库的构建和人类行为在线预测与适应算法的实现。此外，用于安全的人-机器人协作的任务规划算法需要构建从人类演示中学习的运动技能库，并将其与用于在线任务规划和使用所学技能生成目标的算法相关联。用于安全运动规划和控制算法的两层结构包括用于安全HRIS的长期、面向效率的规划层和面向安全的短期控制层。SERoCS将显著扩展协作机器人的技能集，防止和最大限度地减少操作过程中人-机器人碰撞和机器人-机器人碰撞的发生。

In current automated factories, humans and robots typically work separately, partly for safety reasons and partly because full robotic automation has been a goal. In recent years, however, it has been recognized that there are tremendous opportunities when robots are brought out of their cages and allowed to collaborate with human workers in a shared workspace. Such collaboration takes advantage of both the intelligence, adaptability and flexibility of humans and the endurance, strength and reliability of robots. In any collaboration between humans and robots, i.e. co-robots, it is important to consider and ensure both the safety of the humans and the best performance of the robots. This project aims to establish a set of design principles for a safe and efficient robot collaboration system (SERoCS). Outside of factories, SERoCS may be applied in other settings, such as with mobility assistance of humans by robots and automated driving situations where human-driven vehicles and autonomous vehicles share the same road.SERoCS consists of three parts: (1) robust cognition algorithms for environment monitoring, (2) optimal task planning algorithms for safe human-robot collaboration, and (3) safe motion planning and control algorithms for safe human-robot interactions (HRI). Research on cognition environment monitoring algorithms involves the construction of a cognition model library and the implementation of an algorithm for online prediction and adaptation of human behavior. In addition, task planning algorithms for safe human-robot collaboration require the construction of a motion skill library learned from human demonstrations and its association with an algorithm for online task planning and objective generation using learned skills. The two-layer structure that will be employed for safe motion planning and control algorithms comprises a long-term, efficiency-oriented planning layer and a short-term, safety-oriented control layer for safe HRIs. The SERoCS will signicantly expand the skill sets of the co-robots and prevent and minimize occurrences of human-robot collision and robot-robot collision during operation.