To solve the problem of poor flexibility and frequent route breakage caused by dynamic topology in Ad Hoc network routing protocols, a stable ant colony routing algorithm based on Q-learning (SACRQ) is proposed, which synthesizes the Ant Colony Optimization and the Q-learning algorithm. The pheromone level is equal to the Q value to enhance the learning ability of nodes. To avoid local peak, SARCQ applies an adaptive pseudo random proportional action choice rule to select the next hop. A new robustness of the links metric is presented to calculate the probability of the route selection together with the pheromone level. The algorithm enhances the stability of the links and demonstrates high flexibility to the dynamic topology of the network. Simulation results show that SACRQ achieves better performance in terms of the number of the route discovery, the average end-to-end delay, the number of collisions and the average throughput per route discovery, which is respectively compared with the ARA and AODV.