Research On Lateral Balance Control Of Unmanned Bicycle Based On Reinforcement Learning

Throughout the development process of self-balancing control of unmanned bicycles,the use of traditional controllers has achieved good balance control and trajectory control for unmanned bicycles.However,all or part of these controls are limited to more ideal road conditions and environments,when unmanned bicycles are in different road conditions,or walking in an unknown dynamic change environment,the traditional controller can not adapt the unmanned bicycle to the environment,and the parameters need to be readjusted to achieve a stable state.In this paper,a reinforcement learning strategy,a reactive cognitive learning system,is used to study the self-balancing motion control of unmanned bicycles by relying on the interaction between unmanned bicycles and the environment.The combination of reactive cognitive learning system for unmanned bicycles can make up for the shortcomings of traditional control’s weak adaptability to dynamic environment,enhance the environmental adaptability of unmanned bicycles,and achieve stable lateral balance control in changing environments.In this paper,several key problems in the reactive cognitive learning system of unmanned bicycles are studied,and the main research contents are as follows:1)Analyze and select appropriate classification methods,and construct a state classification model for the study of reactive cognitive learning system for unmanned bicycles.The Linear Discriminant Analysis model is used as a classification model for reactive cognitive learning,and the model is evaluated and analyzed.2)Research and improvement of uncertain functions in reactive cognitive learning system.Under the premise that the original uncertainty function decreases with the learning time and the number of learnings,resulting in a gradual decline in learning motivation,this theory improves the uncertainty function,so that the learning motivation is partially improved in the process of decreasing with time,so as to enhance the learning motivation of the system and increase the exploratory nature of the system.3)To study the evaluation matrix,learning coefficients α,and β selection of the reactive cognitive learning system.Set the fixed-parameter evaluation matrix to a dynamically adjustable evaluation matrix;optimize the calculation of the learning coefficient α function,let the value α adjust with the state change,and β select the appropriate value.Comprehensively improve the above learning strategies,realize the optimization of the reactive cognitive learning system of unmanned bicycles,and improve the learning efficiency and learning quality.4)Use MATLAB/Simulink to build a framework of reactive cognitive learning modules for motion control simulation.The Linear Discriminant Analysis model and fuzzy classification were compared and analyzed,and the results prove that the linear discriminant analysis model classification method can make the learning convergence rate faster,the unmanned bicycle rolling angle learning oscillation is smaller,and the convergence range is smaller when it learns to the equilibrium state,which verifies the feasibility of using this method;set different initial expectation values for simulation comparative analysis,and the results prove that the appropriate adjustment of the initial expectation value of the automatic unit can improve the learning efficiency and reduce the learning time,laying the foundation for subsequent physical prototype experiments.5)Using improved reactive cognitive learning,the physical prototype self-balancing learning experiment is carried out.the partial feedback linearization controller was used for physical prototype experiments,and the self-balancing state data of unmanned bicycle based on partial feedback linearization was collected as the basic data of state classification,and the Linear Discriminant Analysis state classification model was constructed;with reference to the simulation experimental process,the experiment of reactive cognitive learning system was carried out on the physical prototype of unmanned bicycle,and experimental results show that the improved system has a longer effective learning time for the self-learning of the lateral balance of unmanned bicycles,the ability to adjust the lateral balance of unmanned bicycles is better,and the convergence range of the rolling angle of unmanned bicycles is closer to the equilibrium range when balancing learning state.This paper explores the application of reactive cognitive learning system to lateral balance motion control in unmanned bicycles.Aiming at several key problems in the reactive cognitive learning system,in-depth analysis and discussion were carried out,and the improved reactive cognitive learning strategy was verified through MATLAB/Simulink simulation,and the feasibility of lateral balance motion control of unmanned bicycles was verified,and then the learning experiment was carried out on the physical prototype of unmanned bicycles to achieve lateral balance motion control of unmanned bicycles.