The field of reinforcement learning (RL) has been receiving significant attention due to the emergence of ambitious endeavours that include automated arm deception, 1v1 Dota, and Atari games. This expansion is consistent with the continued success of supervised deep learning, as demonstrated most significantly by the outcome of the 2012 ImageNet classification event. Neural networks with deep structures have recently been widely used in academics to address challenging issues; including comprehending sophisticated behaviours in changing contexts. As a branch of artificial intelligence, reinforcement learning (RL) presents a viable path towards achieving highly intelligent robotic behaviour. In contrast to supervised learning, which trains networks using labelled datasets, reinforcement learning (RL) is more appropriate for situations in which explicit input is not available since it incorporates experimental encounters with the environment. The fundamental ideas of reinforcement learning and how it is utilised in a variety of fields, such as games for computers, robots, and stock market assessment, are reviewed in this article, along with the different techniques for learning used in multi-agent scenarios. It also goes over how to formulate and solve difficulties related to reinforcement learning, offering perspective on the prospects and difficulties present in this quickly developing discipline.