本文将自行车机器人的研究和机器学习的应用有机结合，以自行车走直线这一简单运动控制问题为例，探究了机器学习在控制领域的应用。以PID控制器为基本框架，先后研究了手调参数的PD控制和机器学习调参的PD控制，后者的研究可为日后的数据驱动控制系统发展做铺垫。研究中为控制实验搭建了实体机，创新地提出双旋翼车把驱动方式，既保持了前轮驱动的灵活性，又避免了对车体的改装。还在机器学习阶段提出了由目标函数性质指导训练数据处理的方法。最终实验结果表明，基于机器学习的PD控制器能够控制小车实现十分稳定的直线行走，抗干扰性能较强。

This research combines the control of the riderless bicycle and machine learning. The case study of how to control a riderless bicycle to drive straight is presented to explore the application of machine learning in the field of automated control. In this research, the PID controller is used as the framework, and PD controller designed by human and PD controller designed using machine learning are studied. The later study will lay the foundation for the development of data-driven controlling in future. In the research, the researcher builds an experimental riderless bicycle and introduces the notion of a driving solution with 2 rotor handlebars, which not only maintains the flexibility of the front-wheel drive but also avoids modifying the body of the bicycle. At the stage of the study of the machine learning PD controller, the approach of processing the training datasets guided by the nature of objective function is applied. The final results of this research demonstrate that the machine learning PD controller can stably control the bicycle to drive straight and resist interferences.