Application Of Adaptive Dynamic Programming And Optimal Output Regulation Theory On Two-Wheeled Wheel-Legged Robots

With the development of robots,two-wheeled wheel-legged robots have attracted much attention due to their fast moving speed and strong passability.Model-based control methods will suffer from degradation of control performance due to modeling deviations,inaccurate model parameters,external interference and other factors.The robot self-learns the optimal controller through a data-driven non-model method,which has certain practical value and significance for its application in actual scenarios.Among the existing two-wheeled wheel-legged robot papers,the Ascento robot paper only introduces the balance control and structural description under a fixed attitude,and has not yet done research on the above two issues.The first generation Ollie robot paper,the stable control domain is extended by designing the controller,but there is no targeted solution to the above two problems.The shortcomings of related research are as follows:The published research results of two-wheeled wheel-legged robots mostly focus on the optimization design of the robot body structure and the artificial deviation compensation for traditional model-based control methods.No systematic method is proposed to solve the problem of control performance degradation caused by inaccurate models and inaccurate model parameters.By consulting a large number of literature and comparing the results of practical applications,the adaptive dynamic programming(ADP)and adaptive optimal output regulation(AOOR)methods are proposed and applied for the first time to the two-wheeled wheel-legged robot Ollie developed by Tencent Robotics X laboratory.Combining these two theories with Ollie’s actual situation,a corresponding self-iterative strategy is designed to solve the above-mentioned two-wheeled wheel-legged robot’s adaptive balance and speed adjustment problems.This method has the advantages of less required data,online data collection and online iteration.Among them,AOOR is a control theory for linear systems.After communicating with the author,it was first applied to two-wheeled wheel-legged robots.These two methods can automatically iterate the optimal controller according to the current real-time state information of the robot,without relying on the precise modeling of the robot and the accurate acquisition of model parameters,and solve the two problems mentioned above.Through four scenarios and a comparative experiment of a balanced rolling cup application,the innovations of this paper are as follows:The data-driven ADP controller allows the robot to maintain an adaptive balance,which solves the problem that the performance of the controller is affected by precise modeling and model parameter accuracy.The data-driven AOOR controller realizes adaptive speed adjustment,allowing the robot to stop in place stably,and no longer needs to manually compensate the robot in different postures.This paper only focuses on the linear system interval when the robot remains upright.How to solve the control problem of the two-wheeled wheel-legged robot outside the linear interval is also a problem that should be considered for the practical application of this type of robot in the future.