A multiagent reinforcement learning control approach to environment exploration

Reinforcement learning (RL) is an area of machine learning and a branch of artificial intelligence (AI). In RL agents freely interact with environment and learn through trial and error with reward and punishment provided by the environment where agents exist. The method is inspired by behaviorist psychology, and similar to how living beings learn by trial and error. An important aspect of RL is that agents aim to maximize end cumulative reward not immediate reward. This is a very important feature as it allows RL to be applied to complex problems where immediate results are very different to final end outcomes. For example, individual moves in a business deal could incur loss, but an overall strategy over time could result in net gain. Such complex problems are well suited to RL. Most RL studies deal with single agent acting in an environment; here we deal with a more efficient paradigm where multiple agents act and cooperate in the environment to maximize the final outcome. A multiagent reinforcement leaning (MARL) can be very effective in finding solutions to complex problems.