We discussed the card game Poker briefly in Section 13.3. Given that the present (probabilistic) state of a player is either good bet, questionable bet, or bad bet, work out a POMDP to represent this situation. You might discuss this using the probabilities that certain possible poker hands can be dealt.

Data from section 13.3

Transcribed Image Text:

In Section 10.7 we first introduced reinforcement learning, where with reward reinforce- ment, an agent learns to take different sets of actions in an environment. The goal of the agent is to maximize its long-term reward. Generally speaking, the agent wants to learn a policy, or a mapping between reward states of the world and its own actions in the world. To illustrate the concepts of reinforcement learning, we consider the example of a recycling robot (adapted from Sutton and Barto 1998). Consider a mobile robot whose job is to collect empty soda cans from business offices. The robot has a rechargeable battery. It is equipped with sensors to find cans as well as with an arm to collect them. We assume that the robot has a control system for interpreting sensory information, for navigating, and for moving its arm to collect the cans. The robot uses reinforcement learning, based on the current charge level of its battery, to search for cans. The agent must make one of three decisions: 1. The robot can actively search for a soda can during a certain time interval; 2. The robot can pause and wait for someone to bring it a can; or 3. The robot can return to the home base to recharge its batteries.