Название: Human-In-the-loop Learning and Control for Robot Teleoperation
Автор: Chenguang Yang, Jing Luo, Ning Wang
Издательство: Academic Press/Elsevier
Год: 2023
Страниц: 260
Язык: английский
Формат: pdf (true)
Размер: 21.9 MB

Robots are utilized more and more frequently nowadays. They are employed in industrial production processes and human daily life with many more promising prospects for applications. As the depth and breadth of robotics applications continue to expand, robots are expected to work in dynamic and unknown environments to meet more demanding requirements for complex and diverse tasks. Remote control of robots with semi-autonomous capabilities, also known as telerobotics, is a popular and well-developed area which has attracted considerable attention from academia and industry. Telerobots are widely used for missions in hazardous environments, because state-of-the-art autonomous robots cannot perform all required by themselves. Many relevant algorithms and techniques have been developed to address the limitations of physical human–robot interaction (pHRI), perception, and learning capabilities in telerobot development. Among these approaches, teleoperation and robot learning have been proven to enhance performance of telerobots with higher stability and efficiency. However, most of these methods focus only on the studies of the characteristics of the telerobots themselves, without taking into account the dominant role and intelligence of human operators in the closed-loop system.

In this book, we will present our recent work on control and learning methods, taking into account the human factors to enable telerobots interacting with humans in a user-friendly and intelligent way. Different aspects of teleoperation control will be investigated in terms of uncertainty compensation, user experience, shared control, and pHRI. Learning plays an important role to help telerobots acquire human manipulation skills, especially in human-in-the-loop teleoperating systems, allowing humans to focus mainly on high-level cognitive tasks such as decision-making. This book will focus primarily on learning and control technologies specifically developed for human-in-the-loop systems that can improve telerobots’ performance. The organization of the book is summarized below:

Chapter 1 will provide an overview of typical remote operating systems. Teleoperating systems include unilateral teleoperation, bilateral teleoperation, and multilateral teleoperation, which can help human operators to accomplish tasks in complex situations. Given the importance of pHRI in teleoperation, we will introduce both unimodal and multimodal interfaces of pHRI that are used to provide human operators with perceptual feedback for more natural and effective interactions with robots. We will also present learning and control algorithms, which will be used to obtain human-like compliant operations. Next, we will present typical examples of telerobot applications such as the da Vinci surgical robot, the Canadian robot, the Kontur-2 project, and Robonaut.

In Chapter 2, we will present the platforms and software systems used for teleoperation in this book. The teleoperation platforms include mobile robots, Baxter robots, KUKA LBR iiwa robots, haptic devices, and different sensors such as Kinect, Mini45 force/torque sensors, and the MYO Armband. The software systems for teleoperation are used to describe and analyze kinematic and dynamic models of robots and various simulation applications. The main robotics software systems include the OpenHaptics Toolkit, MATLAB Robotics Toolbox, the Robotics Operating System, Gazebo, and Coppeliasim.

Chapter 3 will focus on how to control self-driving robots in the presence of uncertainty. Wave variables and neural learning methods show great benefits in dealing with time delays, especially in uncertain systems and environments. In this chapter, basic teleoperation control based on bilateral teleoperation systems, such as position-to-position control and four-channel control, will be introduced. Neural learning control methods will be presented to improve teleoperation performance in the presence of nonlinearity and uncertainty.

Chapter 7 will propose a task learning scheme for a human-in-the-loop teleoperation system for improving the efficiency of robot learning and the automatic generation of tasks. For task learning, the space vector approach and dynamic time warping method are presented to process the demonstration data in 3D space. Then, several robot task trajectory learning methods based on machine learning are proposed, such as the Gaussian mixture model and Gaussian mixture regression, the extreme learning machine, locally weighted regression, and the hidden semi-Markov model, which do not require human involvement after learning human skills, to improve the efficiency. Several experiments are executed successfully to show the performance of our methods.