The highly dynamic topology of ad hoc networks and their limited bandwidth made the routing task more difficult. Actually, various kinds of sources (such as voice, video, or data) with diverse traffic characteristics and Quality of Service Requirements (QoS) are multiplexed at very high rates, leads to significant traffic problems such as packet losses, transmission delays, delay variations, etc, caused mainly by congestion in the networks. The prediction of these problems in real time is quite difficult, making the effectiveness of "traditional" protocols based on analytical models questionable. We propose in this paper a state of the art of this problem followed by a solution based on reinforcement learning paradigm that we find more adapted for this kind of problems.