Beaconing is usually employed to allow network discovery and to maintain synchronisation in mesh networking protocols, such as those defined in the IEEE 802.15.4e and IEEE 802.11s standards. Thus, avoiding persistent or consecutive collisions of beacons is crucial in order to ensure correct network operation. Beacons are also used in receiver-initiated medium access protocols to advertise that nodes are awake. Consequently, effective beacon scheduling can enable duty-cycle operation and reduce energy consumption. In this work, we propose a completely decentralised and low-complexity solution based on learning techniques to schedule beacon transmissions in mesh networks. We show the algorithm converges to beacon collision-free operation almost surely in finite time and evaluate converge times in different mesh network scenarios.