Due to the fact that AGVs used in the field of automatic parking are of great importance and have higher requirements for the smoothness of movement trajectory and walking distance, an improved ant colony algorithm has thus been proposed in view of such flaws as propensity to deadlock, redundance of idle nodes, and uncontrollability of the steering amplitude found in traditional ant colony algorithms. Firstly, by using a map compensation function the map is to be optimized prior to the formal iteration of the algorithm, thus reducing the probability of deadlock. Secondly, the pheromone concentration of the map can be initialized after a map optimization, which helps to accelerate the convergence speed of the algorithm. Finally, by adjusting the path generation logic, the step size can be adaptively adjusted by the proposed algorithm, with the path smoothness improved and the steering swinging reduced. The simulation results show that the improved algorithm reduces deadlock occurrence with a faster convergence speed, a smoother generated path steering, a smaller number of idle nodes and a shorter total path length.