The stochastic dynamics of buses as they serve commuters at bus stops is complex. The ensuing complexity arises from the random interactions of buses and people at the bus stop. If this interaction is left to occur spontaneously, the buses will eventually bunch. To circumvent the consequential transport inefficiency, control protocols have been utilized, with the holding strategy proving to be the most successful. However, the devised protocols typically do not have the adaptive capacity to match the dynamic travel patterns and irregularities of the commuters and ambient traffic conditions. We show that reinforcement learning can potentially bridge this gap through the discovery of new and novel strategies that optimize the system for efficiency within these environments. Interestingly, the uncovered strategies can generate unconventional dynamics between buses and people. In this talk, I will illustrate these dynamics as well as the efficacy of the strategy based on simulation results from an agent-based bus network system.