Aiming at the blindness of randomly generating initial population when genetic algorithm is used to solve robot path planning problem, the initialization algorithm is improved. Firstly, a free grid was randomly selected among the rows between the starting point and the end point to ensure the accessibility of the path. Because the paths composed of these grids were discontinuous, a midpoint connection method was designed to connect the discontinuous points. Finally, the path was simplified to avoid repeated paths. The algorithm was simulated in the same environment as the adaptive genetic algorithm in reference［8］. The experimental results show that the improved genetic algorithm can effectively improve the quality of solution and the speed of evolution.