This paper investigates the routing protocols for randomly distributed wireless sensor networks (WSNs). A learning algorithm is proposed for computing the path on routing,and only few information of the  adjacent nodes is necessary. By using the depth first search strategy,and remembering the historical experience of path on routing,we can find the next node,which illustrates the advantage of combining the  autonomous decentralized systems approach and the  distributed estimation learning algorithm. The simulation results show the usefulness and advantages of the developed algorithm.