As technology progresses, almost all Industrial Internet of Things applications are becoming more data-driven. Hence, more advanced ways of communication are required to both save energy and become efficient to collect huge amounts of data. Researchers have now agreed that the root improvement of IIoT communication can start from a lower layer which is Data Link. Time Division Medium Access-based protocols on this layer have drawn academia's attention to decrease the energy consumption of IIoT devices and achieve higher levels of throughput. Yet, a problem arises, that is, developing a scheduler has always been troublesome while designing a Medium Access Control protocol. Scheduling communication in TDMA-based MAC protocols is considered an NP-hard problem. It is one of the topics highly debated among researchers to make the scheduling adaptive to the huge variation of IIoT application requirements. Conventional scheduling algorithms may suffer from the rapid changes in sensor network topology, network throughput, and applications' delay targets. Therefore, Reinforcement Learning algorithms are being integrated into scheduling techniques because of their adaptability and efficiency. This paper surveys RL-based TDMA MAC protocols and compares them in terms of several unified features. Each protocol is explained and discussed with others regarding its advantages and drawbacks.