Abstract: To overcome the problems of return lag after steering and poor adaptability to bends with varying widths in the reactive navigation of unmanned Load-Haul-Dump (LHD) unit,a reactive navigation system based on non-linear model predictive control (NMPC) is proposed. A motion controller based on NMPC algorithm is designed and a sec- tional local path decision strategy is proposed to match the NMPC controller. A nonlinear predictive model based on LHD model is deduced by Euler method,and an optimal function which consist of limits is set up to solve the optimal control sequence. The sectional local path decision-making strategy divides the underground roadway fitting into several typical working environments and proposes some corresponding decision-making strategies in different environments. The local reference path is adjusted in real time according to the roadway environment information in the predicted time domain during the driving of the LHD and then the local reference path information is provided to the NMPC con- troller for tracking. The reactive navigation system is validated by some simulation experiments in different environ- ments and comparison with the reactive navigation system based on linear time-varying model predictive control (LTV- MPC). The results show that the reactive navigation system proposed can realize an autonomous navigation in narrow roadways,and adapt to the roadways with varying widths. When driving in some different simulation environments,the minimum distance between the LHD unit and the roadway wall is 0. 73 m,and the heading error can converge from 0. 89 rad to stable state which tends to 0. 001 0 rad within 5 s,which greatly improves the safety of the reference path at the bend and the adaptability of the roadway with different widths.