Abstract: At present, non-full section tunnelling is the main way of underground roadway tunnelling in coal mine. Its operation process relies heavily on manual operation, and is affected by the harsh underground environment, limited vision and narrow and long roadway space constraints, which brings challenges to further realizing the automation and intelligence of non-full section tunnelling. The moving control of driving equipment is the core of realizing the autonomous operation of non-full section driving, and solving the problem of moving path planning is the premise of realizing the autonomous moving control. Here, a moving path planning method for non-full section tunnelling with collision forecast is proposed to realize safe path planning of roadheader in narrow roadway space. Firstly, starting from the tunnelling technology, the autonomous driving process is divided into three stages: driving track movement, autonomous shifting and autonomous cutting, forming a complete autonomous operating process of non-full section tunnelling, and developing the overall plan of the path planning of non-full section tunnelling, autonomous shifting. Secondly, according to the autonomous work flow, the path design of the non-full section driving machine is carried out, the mapping model between the cutting position and the cutting forming area is established, and the stopping point of the cutting position is determined. Then, according to the prior information of roadway design, the mathematical model of roadway is constructed, the composite artificial potential field of roadway and roadheader is designed, and the collision prediction model driven by the position and pose of roadheader is established to realize the real-time collision prediction of roadheader. Further, the collision prediction model is introduced to design the segmenting weighted heurism function, and the inflection point of Bezier curve is used to smooth the process, so as to realize the safe path planning of the roadheader. The simulation results show that the proposed method can adapt to different roadway parameters for path point calculation, and the average time of path planning process is 0.135 s. Compared with the original algorithm, the path length is shortened by 0.3271 m on average, the running time is shortened by 0.2462 s, and the path collision does not occur. Finally, a path tracking experiment is built to simulate the tunnel environment to verify the effectiveness of the method. The experimental results show that the trend of the planned path is consistent with the tracking results, and the maximum tracking error is less than 0.056 m, which realizes the safe path planning of non-full section tunnelling. The autonomous moving machine path planning method of non-full section driving combined with collision prediction can effectively solve the path planning problem of narrow roadway space and lay a technical foundation for autonomous cutting.