Deep Predictive Control For Networked Mobile Robots

With the rapid development of artificial intelligence technology,networked mo-bile robots are widely used in various fields such as industrial and agricultural pro-duction,aerospace.unmanned combat,disaster relief,and home services.Because trajectory tracking control is a comprehensive problem that integrates kinematics,in-telligent control,and information processing,it has always been the focus of research on networked mobile robots.Whether it is the trajectory tracking control of the master-slave type mobile robot or the remote trajectory tracking control,the introduction of the communication network breaks the limitation of the traditional control system in the spatial position,and the flexibility of the control system is significantly improved.At the same time,the phenomenon of network data packet loss is unavoidable,and to a certain extent,the performance of the system is degraded or even destabilized.There-fore,it is of great theoretical and practical significance to study the trajectory tracking control problem of networked mobile robots with data packet loss.This paper main-ly studies the problem of data packet loss prediction and compensation in the tracking process of networked mobile robots.The main research work includes the following aspects:1.Given the reference trajectory of the trajectory tracking without considering the acceleration constraints of the mobile robot,there will be cases where the robot cannot reach the given input at some sampling moments,and the trajectory tracking task cannot be completed.To solve this problem,the third-order Bezier curve is used to reconstruct partial curves that do not satisfy the constraints of the acceleration of the robot,so that each point on the path can meet the nonholonomic constraints of the mobile robot.Afterwards,through the planning of the robot time curve.the time opti-mal trajectory under the constraint condition is obtained.Finally,a suitable trajectory tracking control law is designed based on Backstepping control method.2.The trajectory tracking control problem of networked mobile robots is studied.The network packet loss phenomenon may occur in remote control.The characteristics of LSTM-based units are suitable for dealing with problems highly related to time series,as well as their appearance in the fields of speech and images.For the strong ability of sequence data prediction,a data packet loss prediction compensation method based on deep LSTM neural network is proposed.Based on the historical pose and state deviation data of a mobile robot over a period of time,a deep LSTM neural network is used to predict and compensate part of the lost packet time data,ensuring that the mobile robot can still use its predicted value to issue real-time control commands at the time of data packet loss.To complete the effective networked mobile robot trajectory tracking task.Finally,through simulation experiments,this paper firstly proves the feasibility and effectiveness of using the third-order Bezier curve to reconstruct partial paths that do not meet the acceleration constraints of mobile robots.The experimental results show that this method can be used to reconstruct the trajectory curve.Each point satisfies the requirements of mobile robots and speed constraints,and then the trajectory tracking experiment proves the necessity of reconstructing some invalid paths.Secondly,the trajectory tracking of the mobile robot under the traditional packet loss compensation algorithm,artificial neural network algorithm and the depth LSTM neural network pre-diction model proposed in this paper is simulated and compared.The random loss and continuous packet loss experiments are designed respectively.The experimental re-sults verify the results.In this paper,the correctness and effectiveness of the proposed algorithm,and deep LSTM neural network in terms of packet loss prediction and com-pensation have stronger advantages than traditional artificial neural networks,tracking trajectory accuracy is higher,the error convergence speed is faster.In addition,in the presence of system noise,because the deep neural network has a certain role in filter-ing out noise,the trajectory tracking effect of the proposed prediction compensation algorithm is better.