基于机器学习实现双轮机器人平衡控制的应用研究

Research and Application of Balance Control of two-wheeled Robots Based on Machine Learning

为了有效地解决双轮机器人平衡控制问题,改善学习时间过长的问题,使双轮机器人具有自学习能力;将机器学习中强化学习算法应用于双轮机器人上,利用内部回归神经网络构造性能评价函数,设计了一种新型控制算法;该算法是一种不需要预测和辨识模型,在模型未知及没有先验经验的条件下,通过自身神经网络的在线学习,以实现对双轮机器人的自学习平衡控制;Matlab仿真以及物理实验表明:该方法能在短时间内成功实现对双轮机器人的自学习平衡控制,且在性能上优于其它学习算法。

In order to effectively solve the problem of balance control and improve learning efficiency in two--wheeled robot, which makes two--wheeled robot has the self--learning ability, this paper present a novel method to balance control of two--wheeled robot by using reinforcement learning in Machine Learning, and construct performance evaluation function by using internal recurrent neural networks. The learning algorithm is that the model of the robot is not available and the agent has no prior knowledge, that is to say, with no prediction and no identification model. Through their own learning neural network online , it can achieve self--learning balance control in two--wheeled robot. The Matlab simulation and physical experiments demonstrate that it can successfully achieve self--learning balance control of two-- wheeled robot System in a short time, and better than other learning algorithms in performance.