Manipulability ellipsoid on the Riemannian manifold serves as an effective criterion to analyze, measure, and control the dexterous performance of robots. For asymmetric bilateral telerobotics, due to the different structures of leader and follower robots, it is difficult or even impossible for the operator to manually regulate the manipulability of the remote follower robot. Thus, it is desired that the follower robot can automatically regulate its manipulability to assist the operator in remote different task executions, like humans regulating their own postures to enhance manipulability and adapt to different task scenarios. This article proposes a novel framework for manipulability transfer from human to robot. In this framework, we develop a Type-2 fuzzy model-based imitation learning method to encode and reproduce manipulability ellipsoids from demonstrations. This method can achieve high performance in accuracy and computational efficiency. In addition, it supports learning from a single demonstration. Then, we combine this method with a Riemannian manifold-based quadratic programming control algorithm such that the robot manipulability can fast track the desired manipulability profile. This framework is applied to telerobotics, in which a bilateral teleoperation controller is designed that enables the robot to follow the operator's command and simultaneously self-regulate its manipulability to perform the task adaptively. Meanwhile, the operator can receive force feedback relating to the manipulability regulation. Evaluations using comparative studies and practical experiments with a 3-degrees of freedom (DoF) haptic device and 7-DoF robots are presented to show the effectiveness of the proposed framework.

中文翻译：

---

机器人操纵性学习与控制框架及其在远程机器人中的应用


黎曼流形上的可操纵性椭球是分析、测量和控制机器人灵巧性能的有效标准。对于非对称双边遥控机器人，由于引导机器人和跟随机器人的结构不同，操作者很难甚至不可能手动调节远程跟随机器人的操控性。因此，希望跟随机器人能够自动调节其可操作性，以协助操作者远程执行不同的任务，就像人类调节自己的姿势一样，以增强可操作性并适应不同的任务场景。本文提出了一种从人类到机器人的可操作性转移的新颖框架。在此框架中，我们开发了一种基于 Type-2 模糊模型的模仿学习方法，用于从演示中编码和再现可操作性椭球体。该方法可以在精度和计算效率方面取得较高的性能。此外，它还支持通过单个演示进行学习。然后，我们将该方法与基于黎曼流形的二次规划控制算法相结合，使机器人的可操纵性能够快速跟踪所需的可操纵性曲线。该框架应用于远程机器人，其中设计了双边远程操作控制器，使机器人能够遵循操作员的命令，同时自我调节其可操作性以自适应地执行任务。同时，操作员可以接收与可操纵性调节相关的力反馈。通过使用 3 自由度 (DoF) 触觉设备和 7-DoF 机器人的比较研究和实际实验进行评估，以显示所提出框架的有效性。