Balancing and walking control of biped robots using contact wrench based on whole-body control framework

Abstract

Since biped robots can walk and operate in a human environment, the need for disaster response or service fields is increasing. While the human environment is complex and irregular, there is limit to the recognition and control precision of the robot. Cognitive errors can lead to unexpected contact between the robot and the environment. In case of unexpected contact, whole-body controller based on joint torque control provides compliant motion control or force control scheme. Therefore, the robot receives a relatively small reaction force through contacts, so it is possible to control the joint torque robot more safely than joint position control methods. In this research, the operational space based whole-body control framework is extended to consider contact force distribution for biped robot. Then, the framework is used to develop three application methods to keep the robot walking without losing balance when unintentional contact occurs. The first is a force generation method for the high-velocity center of mass (CoM) motion during dynamic walking of the robot. The existing method for the high-velocity CoM motion control is focused on the trajecotry generation method, but it is not suitable for the torque control based biped robot controller in which high performance trajectory tracking is not guaranteed. The proposed method can be used even if the disturbance acts because it considers only the force for the maximum acceleration and deceleration that can be generated according to the state of the robot without the time-dependent trajectory planning. The robot balance control, which uses the robot dynamics and the null-space to control the balance is also developed. Through the balance controller, it is possible to control the robot not to fall even when a large disturbance acts on the robot. Finally, an active sensing method is also developed to estimate the edge where the contact between the robot and the terrain occurred. This is a creative approach in that it can correct erroneous recognition information using contact information rather than controlling the robot to escape from unexpected contact. By using these three developed methods, the robot can be expected not to fall over even if the robot is in conflict with the terrain due to the perception error, but to continue walking after searching again for the correct terrain. All of the developed methods utilize robot dynamics information and directly generate torque in the joints, so they can be used without the help of external sensors such as F/T sensors. The results of the study are verified by experimenting with 12-DOF torque control biped robot DYROS RED, and the process of implementing operational space based whole-body control framework in the robot is also explained in this paper.