Interaction Dynamics and Control of Collaborative Industrial Mobile Robot Manipulators for SME Manufacturing

TY - BOOK

T1 - Interaction Dynamics and Control of Collaborative Industrial Mobile Robot Manipulators for SME Manufacturing

AU - Zhou, Zhengxue

PY - 2022/11

Y1 - 2022/11

N2 - Collaborative industrial mobile manipulators (CIMMs), consisting of collaborative robot arms and mobile platforms, offer the advantages of both high mobility and manipulability. Compared to conventional stationary robot manipulators, CIMMs are more adaptive to dynamic environments and have better potential applications for collaborating with humans and environments. CIMMs provide the potential solution to facilitate automatic production in small and medium enterprises (SMEs) in which robots are required to collaborate/interact highly with human workers due to the limited space in SMEs. With the aim of human-robot collaboration, this dissertation solves several challenges relevant to coupling dynamics, safe interaction, work piece detection, and the complexity of learning and training. This dissertation is organized by following the solutions to the challenges addressed below. Currently, CIMMs are often operated at reduced speeds to achieve safe physical human-robot interaction and avoid hazards from collision, but it limits the production efficiency of robots. Increasing operation speed while ensuring safe interactions requires a CIMM to generate accurate torques/forces, thus causing the need for a precise dynamic model. However, the coupled kinematics and dynamics between the robot arm and the mobile platform are the major challenges for establishing a precise dynamic model of a high-Degree of Freedom (DOF) robot, such as a CIMM. In addition to the challenge of establishing an accurate dynamic model, the safety issues of the physical interaction between CIMMs and humans or environments are another concern while executing manipulation tasks by robots. The interaction relationship may significantly vary during the task execution due to the uncertain contact dynamics and friction. The uncertainties from the dynamic interaction lead to the difficulties of controlling robots to execute manipulation tasks stably and precisely. The changeable and unstructured environments in SMEs present another challenge for navigating a multitasking CIMM among different working stations. The conventional navigation approach, i.e., simultaneous localization and mapping (SLAM), suffers from the problem of unstable performance in changeable environments. Consequently, one of the critical challenges for robot navigation in SMEs is developing a robust detection algorithm for a CIMM against the changeable layout. Calculating the precise relative position of detected objects for the CIMM is another challenging issue. Machine learning is a promising method to solve the challenges robots face while executing manipulation tasks and navigating in unstructured environments. However, acquiring sufficient datasets to train or teach a robot to navigate or manipulate robustly is time-consuming. In addition, training and evaluating contact-rich tasks or human-robot interaction tasks are extremely challenging since a robot learns by exploring the environment randomly, which is hazardous for the robot and its surroundings. Besides, the time cost is another challenging issue since training a robot for a manipulation task requires lots of steps or episodes, and it would be time-consuming for real and physical robot training. Finally, various research efforts have been undertaken to explore the application of CIMMs to achieve production tasks in factories. However, most of the efforts were on simulation work or experimental testing conducted in laboratory environments to simulate production tasks using CIMMs. This dissertation aims to propose state-of-the-art techniques for solving the above challenging problems and promoting the application of CIMMs in SME production efficiently and safely. The first goal is to develop a coupled dynamic model of the CIMM as a basis for human/environment-robot interaction. Secondly, variable impedance control is proposed to control the interaction between the CIMM and the environment in a contact-rich task based on imitation learning. Thirdly, learning-based 3D detection incorporating the 2D eye-hand calibration is developed to enable the CIMM to automatically navigate, charge, and avoid humans in an unstructured SME environment. Fourthly, the dynamic model integrates into a three-layer digital twin to represent the joint torques virtually. Furthermore, another digital twin with precise contact dynamics in task space is proposed for training robot manipulation tasks. Finally, the relevant experimental validation and onsite demonstrations of the above-proposed techniques are presented. This dissertation highlights the importance of artificial intelligence (AI) and robotic techniques in facilitating automatic production in SMEs. The proposed robotic solution also provides insights into other fields that need mobile manipulation.

AB - Collaborative industrial mobile manipulators (CIMMs), consisting of collaborative robot arms and mobile platforms, offer the advantages of both high mobility and manipulability. Compared to conventional stationary robot manipulators, CIMMs are more adaptive to dynamic environments and have better potential applications for collaborating with humans and environments. CIMMs provide the potential solution to facilitate automatic production in small and medium enterprises (SMEs) in which robots are required to collaborate/interact highly with human workers due to the limited space in SMEs. With the aim of human-robot collaboration, this dissertation solves several challenges relevant to coupling dynamics, safe interaction, work piece detection, and the complexity of learning and training. This dissertation is organized by following the solutions to the challenges addressed below. Currently, CIMMs are often operated at reduced speeds to achieve safe physical human-robot interaction and avoid hazards from collision, but it limits the production efficiency of robots. Increasing operation speed while ensuring safe interactions requires a CIMM to generate accurate torques/forces, thus causing the need for a precise dynamic model. However, the coupled kinematics and dynamics between the robot arm and the mobile platform are the major challenges for establishing a precise dynamic model of a high-Degree of Freedom (DOF) robot, such as a CIMM. In addition to the challenge of establishing an accurate dynamic model, the safety issues of the physical interaction between CIMMs and humans or environments are another concern while executing manipulation tasks by robots. The interaction relationship may significantly vary during the task execution due to the uncertain contact dynamics and friction. The uncertainties from the dynamic interaction lead to the difficulties of controlling robots to execute manipulation tasks stably and precisely. The changeable and unstructured environments in SMEs present another challenge for navigating a multitasking CIMM among different working stations. The conventional navigation approach, i.e., simultaneous localization and mapping (SLAM), suffers from the problem of unstable performance in changeable environments. Consequently, one of the critical challenges for robot navigation in SMEs is developing a robust detection algorithm for a CIMM against the changeable layout. Calculating the precise relative position of detected objects for the CIMM is another challenging issue. Machine learning is a promising method to solve the challenges robots face while executing manipulation tasks and navigating in unstructured environments. However, acquiring sufficient datasets to train or teach a robot to navigate or manipulate robustly is time-consuming. In addition, training and evaluating contact-rich tasks or human-robot interaction tasks are extremely challenging since a robot learns by exploring the environment randomly, which is hazardous for the robot and its surroundings. Besides, the time cost is another challenging issue since training a robot for a manipulation task requires lots of steps or episodes, and it would be time-consuming for real and physical robot training. Finally, various research efforts have been undertaken to explore the application of CIMMs to achieve production tasks in factories. However, most of the efforts were on simulation work or experimental testing conducted in laboratory environments to simulate production tasks using CIMMs. This dissertation aims to propose state-of-the-art techniques for solving the above challenging problems and promoting the application of CIMMs in SME production efficiently and safely. The first goal is to develop a coupled dynamic model of the CIMM as a basis for human/environment-robot interaction. Secondly, variable impedance control is proposed to control the interaction between the CIMM and the environment in a contact-rich task based on imitation learning. Thirdly, learning-based 3D detection incorporating the 2D eye-hand calibration is developed to enable the CIMM to automatically navigate, charge, and avoid humans in an unstructured SME environment. Fourthly, the dynamic model integrates into a three-layer digital twin to represent the joint torques virtually. Furthermore, another digital twin with precise contact dynamics in task space is proposed for training robot manipulation tasks. Finally, the relevant experimental validation and onsite demonstrations of the above-proposed techniques are presented. This dissertation highlights the importance of artificial intelligence (AI) and robotic techniques in facilitating automatic production in SMEs. The proposed robotic solution also provides insights into other fields that need mobile manipulation.

M3 - Ph.D. thesis

BT - Interaction Dynamics and Control of Collaborative Industrial Mobile Robot Manipulators for SME Manufacturing

PB - Århus Universitet

ER -