/ Research Highlight

Constrained Disturbance Observer

October 11, 2024
In robotics, achieving accurate motion tracking is a basic and fundamental challenge. For accurate tracking, robust control methods such as the Disturbance Observer (DOB) have been studied to eliminate uncertainties such as external forces and joint friction. However, the need for safe interactions has grown as robots increasingly share workspaces with humans. This creates a paradox: robots must not only track accurately but also respond safely to external forces that occur by human interactions. To address this, Prof. Min Jun Kim’s group at KAIST developed the Constrained Disturbance Observer (CDOB) framework, which adds intelligence to the ordinary DOB through optimization techniques. Consequently, while maintaining accuracy during free motion, robots are also capable of managing safety constraints and interacting with unknown environments.

In robotics, accurate tracking has been fundamentally studied in the robust control field for a long time. Such research is aimed at eliminating all uncertainties, including external forces acting on the robot and friction, to ensure precise performance. However, as the concept of robots sharing workspaces with humans has emerged, the need for safe interaction has become increasingly important. This presents a paradox: while precise tracking is crucial, robots also need to respond to external forces (e.g., human interaction) to ensure safety.

Traditionally, this problem has been addressed using additional sensors such as force sensors. However, this issue remains unresolved in sensorless settings, which offer an important cost advantage in industrial applications. To tackle this challenge, Prof. Min Jun Kim’s group at KAIST proposed a novel approach called the ‘Constrained Disturbance Observer (CDOB)’ framework. This new framework extends the disturbance observer’s (DOB) capabilities, a robust control technique for accurate tracking, and enables the robot to exhibit versatile behaviors through optimization techniques.

The following figures show CDOB examples of robotic manipulators. Figure 1 illustrates how the robot’s contact responsiveness is enhanced using the CDOB. Unlike the ordinary DOB, these added properties enable the robot to adapt safely to unknown environments by adjusting its movements during contact. Figure 2 further highlights this property by demonstrating how the robot applies a gentler force to an egg. This ability to regulate force through CDOB showcases a new level of control and safety. For additional details, readers are encouraged to view the attached video.

Figure 1. The robot complies with the unknown obstacle obstructing the trajectory.
Figure 1. The robot complies with the unknown obstacle obstructing the trajectory.
Figure 2. Enable contact-responsive motion (Left). Ordinary DOB results in excessive torque (Right)
Figure 2. Enable contact-responsive motion (Left). Ordinary DOB results in excessive torque (Right)

These behaviors are possible because the CDOB sacrifices some of its robustness by applying optimization techniques. This optimization allows for various constraints, such as those related to safety, inputs, or the robot’s position and velocity. Under normal conditions, the CDOB behaves just like the standard DOB, providing precise control as long as the constraints are satisfied. However, when the system is close to violating these constraints, the CDOB adjusts its behavior to maintain as much accuracy as possible within the given constraints.

For technical details, please refer to the published papers: (i) J. W. Han, D. Park, and M. J. Kim, “Constrained Nonlinear Disturbance Observer for Robotic Systems” in IEEE International Conference on Robotics and Automation (ICRA), Yokohama, Japan, 2024, pp. 3526-3532. (ii) T. H. Yun and M. J. Kim, “Disturbance Observer With Constraints” in IEEE Control Systems Letters, vol. 8, pp. 1949-1954, 2024.

This work was supported in part by a National Research Foundation of Korea (NRF) grant funded by the Korean Government (MSIT) under grant 2021R1C1C1005232 and No. 2021R1A4A3032834; by Samsung Electronics Company Ltd. under grant IO220816-01989-01; by the ITECH Research and Development Program of MOTIE/KEIT under grant 20014398; and by the Challengeable Future Defense Technology Research and Development Program (No.915102201) of Agency for Defense Development in 2024.

Contact Information:
Prof. Min Jun Kim, Ph.D candidate Ji Wan Han, MS candidate Tae Ho Yun Dept. of Electrical Engineering, KAIST
E-mail: minjun.kim@kaist.ac.kr
Homepage: https://sites.google.com/view/kaist-roboticslab
Back to Top