This paper studied the stable balance walking control problem of unmanned bicycles. By simplifying the unmanned bicycles as four rigid bodies and considering the motion characteristics of each body, an underactuated nonlinear dynamics model is obtained using Lagrange equation method. To address the underactuated nature of the system, an underactuated sliding mode controller (USMC) is designed using matrix transformation techniques. The balance control of the unmanned bicycle is achieved by controlling the handlebar using feedback information of handlebar steering angle and roll angle. The parameters of the USMC were determined based on the principles of Lyapunov stability theory and Hurwitz stability criterion, and extensive simulation experiments were conducted to evaluate its performance. The simulation experiments show that the USMC has faster convergence than the PD control algorithm and it can achieve steady straight balance walking of the unmanned bicycle on the ground with good robustness and disturbance rejection.