Abstract This paper addresses the problem of human grasp position estimation in a physical human–robot object handling scenario. The problem is formulated as a linear regression by considering the human grasp position and their exerted torque as unknown parameters. We propose a modified least-squares algorithm to estimate the parameters by evaluating the quality of the estimates based on the assumption that the parameters should remain constant for a period of time. The solution is model-agnostic in terms of the human force/torque model – requiring only force/torque measurements on the robot side and proprioception – and is model-based in terms of the object model. The proposed grasp position estimation method is compared statistically with a conventional contact point estimation method using the collected experimental data. Moreover, the performance of the developed method is evaluated through various scenarios of physical human–robot interaction.

中文翻译：

---

用于人机协作对象操纵的人体抓取位置估计

摘要 本文解决了在物理人机对象处理场景中人类抓取位置估计的问题。通过将人类抓握位置及其施加的扭矩视为未知参数，该问题被表述为线性回归。我们提出了一种改进的最小二乘算法，通过基于参数应该在一段时间内保持恒定的假设来评估估计的质量来估计参数。该解决方案就人类力/扭矩模型而言是模型不可知的——只需要在机器人侧和本体感觉上进行力/扭矩测量——并且在对象模型方面是基于模型的。使用收集的实验数据，将所提出的抓取位置估计方法与传统的接触点估计方法进行统计比较。而且，