Reinforcement learning has a ‘Environment’ which is the problem set we are trying to solve and an ‘Agent’ which is our AI algorithm. The relation between Environment and Agent is really simply. The Agent will perform certain action (like the moving taxi up, down, right, left etc in our toy code) and as a result of this action his state will change (new position of taxi) which will lead to Agent getting a reward (positive reward for reaching its destination, or negative if you incorrectly pickup or drop a passenger). By iterating through action and reward process, Agent learns the Environment. It comes to understand in which state, which action gives him maximum reward and which give him negative reward. This process of performing an action and learning from the reward is called Reinforcement learning.
Ali's Researches 