Motivation behind RL

../_images/motivation.png

Fig. 1 Phases of Industrial Revolution

Industrial Revolution has played a pivotal role in the development of technology and services. In simplest sense, it can be defined as automation of repeated physical solutions.

It has paved way into automation of repeated mental solutions, forming the focus of Digital revolution, which led into the Information age.

../_images/robot_think.jpg

Fig. 2 Next Step: Allow machines to find solutions themselves

Reinforcement Learning is one of the pieces of the puzzle that fits into the next step, i.e., learning autonomously how to make decisions. Under this, we can point out at least two distinct reasons to learn,

  1. Finding previously unknown solutions
    For example,
    • Achieving human level performance in atari games
    • Finding shortest or optimal paths in a network or maze
    • Finding optimal decision making strategy for any custom system with very high dimensional spaces
  2. Finding solutions online during operation
    For example,
    • robot learning to navigate terrains which greatly differ from expectations
    • A cleaning bot learning the best strategy to clean any room or house that it is placed into
    • Learning optimal control decisions like increasing or decreasing temperature for an automatic thermostat