In 2015, Google’s DeepMind AlphaGo made big news in the AI world using deep reinforcement learning. AlphaGo beat a human professional in the very complex game of Go. We’ve seen computers win against before those. So, what was so special this time? During the game, AlphaGo used original moves that it had come up with on its own. Moves that taught the world new knowledge about an ancient game. That’s because AlphaGo optimized its own moves based on the goal of winning rather than say looking at possible moves and choosing a likely response based on a library of examples. In robotics, deep reinforcement learning can also be used to find the optimal solution to a problem. The Japanese company Fenix, uses deep RL to reprogram robots to perform new tasks. Traditionally, an industrial robot needs to be precisely programmed in a tightly controlled environment in order to do something like grasp an object. This is expensive and time-consuming. Fenix robot uses deep RL to train itself to perform a new task. It tries picking up objects while capturing video footage of the process. Each time it succeeds or fails, it refines the deep neural network that controls its action. Warehousing companies that already use robots are looking at deep RL to lower costs. Companies like Amazon with fulfillment centers that deploy thousands of robots are highly motivated to use innovative technologies like deep RL to optimize their resources and remain competitive in the marketplace. A number of companies including big companies such as Google, Autodesk, and Amazon as well as a myriad of smaller companies and startups, are actively working on deep RL solutions in robotics.