IL is the art of acquiring a new skill by emulating an expert. This property of learning from imitation is not strictly necessary for learning sequential decision-making policies but nowadays, it is essential in plenty of problems. Some tasks cannot be solved through mere reinforcement learning, and bootstrapping a policy from the enormous spaces of complex environments is a key factor. The following diagram represents a high-level view of the core components involved in the imitation learning process:
If intelligent agents (the experts) already exist in an environment, they can be used to provide a huge amount of information to a new agent (the learner) about the behaviors needed to accomplish the task and navigate the environment. In this situation, the newer agent can learn much faster without the need to learn from scratch. The expert agent can also...
