This paper proposes a dynamic path planning method in order to solve the problem of traveling path planning of partial section cantilever roadheader and realize unmanned tunneling of roadheader. This method is based on variation self-adaptive particle swarm optimization algorithm(VSPSO) and action-grid(A-G). Based on the analysis of roadheader traveling characteristics and coal mine roadway characteristics, a grid network model and a cost model based on behavior rules are established. By setting the type of cost function and the range of consumption coefficient in the model, and based on the roadway simulation data, we complete a comparative analysis of the proposed VSPSO algorithm and six improved PSO algorithm in the roadheader path planning model. The experimental results show that under the test function, the VSPSO algorithm converges faster and converges more accurately. The VSPSO algorithm has the highest convergence speed and accuracy under the behavior rule grid network model, and can plan the optimal traveling path according to the behavior characteristics of roadheader.