Abstract
Most of the controllers for quadruped robots are based on the planning of their centers of mass (CoM), with the assumption that the CoM locates at the geometrical center of the robot trunk. However, this assumption rarely meets the fact of the robot prototype and introduces apparent influences to the system. This article proposes a CoM estimation and adaptation method for quadruped robots in dynamic trot gait based on the model predictive control (MPC) of the trunk. The influences of the CoM offset on the robot states are inherently analyzed. The CoM components on the horizontal plane are taken into consideration and estimated based on the speed errors and the contact force differences, which are compensated to the robot controller. In the case of payload changes, the identification method for the mass of payload is proposed based on the estimated CoM. Simulation results verify that with the proposed approach, the robot successfully adapts to the CoM payload changes while the locomotion is not interrupted, and the state errors are significantly reduced.
Highlights
Many controllers for legged robots [1] [2] are based on their dynamic models [3], to which the inertial parameters are of great importance, especially for those performing high dynamic locomotion [4]
The robot runs in trot gait with the duty factor of 0.5 and the step frequency of 1.5 Hz, which means that there are two and only two diagonal legs in contact with the ground at any moment, while the other two legs are in the air
In order to evaluate the accuracy of the center of mass (CoM) detection results, the CoM of the robot is set to be at the position rcom = [0.15m, −0.08m, 0.0m]T with respect to the base frame, which is regarded as the real value
Summary
Many controllers for legged robots [1] [2] are based on their dynamic models [3], to which the inertial parameters are of great importance, especially for those performing high dynamic locomotion [4]. Author et al.: Preparation of Papers for IEEE TRANSACTIONS and JOURNALS [17] designed a state estimator for the humanoid robot in the presence of unknown modelling errors, but did not allow the COM offset and the external force disturbances simultaneously. Muscolo [18] presented an approach to minimize the CoM error between the ideal virtual model and the real platform of the humanoid robot using force/torque sensors mounted on the feet, but did not introduce the reflecting method for payload changes. Focchi [19] designed a method for the quadruped robot HyQ [20] to estimate the CoM position offset from its CAD model by collecting numerous sets of sensor feedback in static poses. To overcome the shortcomings above, this paper devotes to the adaptive control method for quadruped robots with unknown CoM offsets and payload changes.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
