Aiming at the problem that some key dynamical states of tractor semi-trailer cannot be measured and the values of sensors are interfered by random factors such as engine vibration noise， an improved particle filter is proposed to estimate the dynamical states of the driving tractor semi-trailer in real-time. This paper establishes a 17 degrees of freedom dynamical model of tractor semi-trailer first. By combining the particle filter principle and the adaptive genetic algorithm to enhance the particle diversity， the piecewise proposal distribution function is designed， and the systematic resampling method is used to suppress the particle regression. The in-time and accurate estimation of longitudinal speed， lateral speed， yaw rate， and other states of tractor semi-trailer was realized. A hardware-in-the-loop （HIL） simulation test platform was built to verify the algorithm under different conditions. The testing results show that compared with the unscented particle filter algorithm， the improved particle filter algorithm proposed in this paper can realize the state estimation of the whole vehicle under both ideal and random noise environments， and has higher estimation accuracy.