Reinforcement learning is a computational approach to understanding and automating goal-directed learning and decision making. RL techniques solve problems through an agent, which acquires experiences through interactions (trial and error) with a dynamic environment. The result is a policy that can resolve complex tasks without specific instructions on how the tasks are to be achieved. In other terms, reinforcement learning can be said to be a computational approach towards learning through interactions (behavioral psychology) that is applied by humans in nature where we learn from the mistakes committed and try to not perform the same mistake again when a similar situation arises. Reinforcement learning has better generalizing properties and differs from supervised learning, which uses labeled examples- because labels might not be representative enough to cover all situations. Unsupervised learning is typically about finding structure hidden in collections of unlabeled data and thus differs from reinforcement learning.
RL problems assume that an optimal reward function is given
and build on it to form a policy for the agent. Reward function is the most succinct representation of the users intention since it specifies the intrinsic desirability of an event for the system. But, providing a reward function is a non-trivial problem and can lead to major design difficulties. Inverse Reinforcement Learning (IRL) is more helpful in such cases, where the reward function is learned through expert demonstrations. In the recent years, IRL has attracted several researchers in the communities of artificial intelligence, psychology, control theory, and machine learning. IRL is appealing because of its potential to use data recorded in everyday tasks (e.g., driving data) to build autonomous agents capable of modeling and socially collaborating with others in our society a form of transfer learning. IRL is also an important approach for learning by demonstration in various settings including robotics and automatic driving. Some applications where IRL has been successfully used are Quadruped locomotion, Helicopter Aerobatics, Parking lot navigation, Urban navigation.