The application of model predictive control in the path tracking control of mobile robots is increasingly widespread, and the real-time performance of controllers is gradually being noticed. At present, the common real-time optimization scheme is a linearizing prediction model scheme, which converts the nonlinear model predictive control into the linear time-varying model predictive control. However, the linearizing prediction model scheme will weaken the ability of the controller to respond to sudden changes in the curvature and heading of the reference path. Therefore, from the nonlinear model predictive control, two real-time optimization schemes were proposed, namely reducing the number of control steps or reducing the control frequency. In the results of simulation and experiment, in each control period, the calculation time of the nonlinear model predictive controller, which was optimized by reducing the number of control steps or reducing the control frequency, was shorter than the control period. At the same time, it can be known from the simulation and experiment that the scheme of reducing the number of control steps had smaller displacement errors and heading errors than the scheme of reducing the control frequency or the linearizing prediction model. That was, adopting the scheme of reducing the number of control steps can better ensure the control accuracy when tracking the reference path which with rapid changes in curvature and heading. Therefore, the scheme of reducing the number of control steps was more suitable than other real-time optimization schemes, for mobile equipment such as agricultural robots that require higher flexibility.