사족 로봇의 중심역학 기반의 몸통 자세 제어 (Body Posture Control for Quadruped Robots Based on Centroidal Dynamics)

서지정보

· 발행기관 : 국방로봇학회
· 수록지 정보 : 국방로봇학회 / 3권 / 2호 / 6 ~ 14페이지
· 저자명 : 김경환, 김정엽

초록

This paper proposes a centroidal dynamics-based body posture control method for a quadrupedal walking robot to maintain balance. To control the body posture as desired, the current state of the robot is estimated through a Kalman filter, with the estimated state defined as the position and velocity of the body center in the ground coordinate system, as well as the foot position. The input to the Kalman filter is the linear acceleration obtained from the IMU attached to the body, and the measurements are set as the foot positions and velocities in the leg coordinate system calculated through joint encoders and kinematics. Firstly, for body posture control of the robot, the weight of the significantly lighter legs compared to the robot body is ignored, and the robot body is assumed as a single rigid body, defining a Centroidal Dynamics model suspended in the air. The reference forces to be applied to the body based on Centroidal Dynamics are calculated through the error between the reference body posture and the actual body posture, and a cost function minimizing the difference between the actual and reference body forces generated by the ground reactions at the feet is formulated, subject to the constraint of limited foot friction, using Quadratic Programming (QP). The optimal ground reaction forces at the feet are derived through QP and the joint torques of the legs are then calculated to generate to perform the body posture control of the quadruped walking robot. The effectiveness of this body posture control method is validated through dynamic simulations of the quadruped robot, PongBot.

영어초록

This paper proposes a centroidal dynamics-based body posture control method for a quadrupedal walking robot to maintain balance. To control the body posture as desired, the current state of the robot is estimated through a Kalman filter, with the estimated state defined as the position and velocity of the body center in the ground coordinate system, as well as the foot position. The input to the Kalman filter is the linear acceleration obtained from the IMU attached to the body, and the measurements are set as the foot positions and velocities in the leg coordinate system calculated through joint encoders and kinematics. Firstly, for body posture control of the robot, the weight of the significantly lighter legs compared to the robot body is ignored, and the robot body is assumed as a single rigid body, defining a Centroidal Dynamics model suspended in the air. The reference forces to be applied to the body based on Centroidal Dynamics are calculated through the error between the reference body posture and the actual body posture, and a cost function minimizing the difference between the actual and reference body forces generated by the ground reactions at the feet is formulated, subject to the constraint of limited foot friction, using Quadratic Programming (QP). The optimal ground reaction forces at the feet are derived through QP and the joint torques of the legs are then calculated to generate to perform the body posture control of the quadruped walking robot. The effectiveness of this body posture control method is validated through dynamic simulations of the quadruped robot, PongBot.

참고자료

· 없음