The advent of collaborative robotics in industry has created a closer collaboration between humans and robots. This has led to the need to optimally schedule human and robot tasks to be robust enough to handle variability induced by time-related operator errors caused by the inability to accurately forecast the stochastic nature of human behavior. This article proposes an explicit scheme for tackling time-related variability in human tasks online in applications where humans intervene at a given time in the collaborative workspace. The planning problem is reformulated as a Travelling Salesman Problem combined with a 0/1-Knapsack Problem in order to actively define robot behavior when there is an unmodelled shift in the human execution time sequence. The method uses a two-level adaptation scheme. The first one (offline) inputs the predicted human behaviour in terms of time required for different activities at each work cycle, and then computes an overall task schedule to minimize the robot's operation time and idle time. The second one (online) involves the real-time detection of the human's timing to either stop the prescribed plan or enhance it in order to minimize robot and human idle times, thereby optimizing the sense of ease and fluency in the interaction. The system is simulated in different scenarios where the human predicted time is set to be wrong, and thus the system needs to account for such variation. The effect of the human predicted time on the task schedule is presented and helps to demonstrate the effectiveness of the proposed approach in dealing with human variability without prior modeling knowledge of the human task time distribution.

中文翻译：

---

用于在制造操作中与人类协作的鲁棒动态机器人调度

工业界协作机器人的出现使得人类和机器人之间的合作更加紧密。这就需要对人类和机器人任务进行最佳调度，使其足够鲁棒，以处理由于无法准确预测人类行为的随机性而导致的与时间相关的操作员错误所引起的变化。本文提出了一种明确的方案，用于在人类在协作工作空间中的给定时间进行干预的应用程序中解决在线人类任务中与时间相关的变化。规划问题被重新表述为旅行商问题与 0/1 背包问题相结合，以便在人类执行时间序列中存在未建模的转变时主动定义机器人行为。该方法使用两级自适应方案。第一个（离线）根据每个工作周期不同活动所需的时间输入预测的人类行为，然后计算总体任务计划，以最大限度地减少机器人的操作时间和空闲时间。第二个（在线）涉及实时检测人类停止或增强规定计划的时机，以最大限度地减少机器人和人类的空闲时间，从而优化交互中的轻松感和流畅感。该系统是在不同的场景下进行模拟的，其中人类预测的时间设置是错误的，因此系统需要考虑这种变化。提出了人类预测时间对任务调度的影响，并有助于证明所提出的方法在处理人类变异性方面的有效性，而无需事先了解人类任务时间分布的建模知识。