Research On Self-adaptive Adjustment Method Of Mobile Robot Control Model

Executing tasks in an open-environment,the mobile robot is controlled with a control model based on the cognition of environmental changes.Caused by the open-environment,the mobile robot is controlled with following challenges:(1)The uncertainty of the open environment makes the prior knowledge of model design missing,which leads to unknown risks in robot behavior decision-making;(2)Limited by physical environment and physical behavior,the execution effect of mobile robot behavior decision-making does not completely match expectations;(3)The existence of hysteresis of sensed information requires mobile robots to generate behavior decisions in a short time to avoid sensed information failure.To tackle above challenges,this paper proposes a self-adaptive adjustment approach for robot control models based on empirical data learning.Firstly,to handle the openness of the environment,the linear regression method is designed to learn how environmental changes affect the impact factors of robot control model(such as the P-value in the PID model).Secondly,according to the environmental effects on the robot task execution,the density-based clustering algorithm is applied to divide the training dataset into several sub-dataset.Learning from the clustering results,the mutual interference between training data can be avoided.Thirdly,to deal with the hysteresis of sensed information,real-time feedback control loop is combined with off-line learning result in this paper.Particularly,the learning processes that require a lot of time and overhead are placed offline.During the real-time control of the mobile robot,off-line learning results and environmental changes are organized to adjust the impact factors.In this way,the control model will be able to dynamic update based on environmental changes.To verify that the proposed approach has strong adaptive ability to uncertain environmental changes in an open environment,the paper carried out two experiments,including simulation experiments based on V-REP simulator and physical experiments based on Turtlebot.Firstly,in order to evaluate that the proposed approach has the ability to respond to different or even unknown environmental changes,this paper designs a Path-following case for motorcycles in the simulation experiment.Compared with the DQN and MPC methods,the proposed approach responds more quickly to environmental changes and the behavioral deviation of the robot is smaller.Secondly,in order to evaluate the feasibility of the proposed approach,the physical indoor robot Turtlebot used the purposed approach to complete the Object-following case in a completely unknown complex outdoor environment.