Q Learning Properties

• still need - iterative approximation or recursive definition
• , so
• , the learner's estimate of , is stored in a big table which is Initially filled with random values or zero
• The agent starts in some state, , and chooses some action, , and observes the result reward, , and the new state,
• It then updates the table,
• doesn't need to know functions or just executes the action and observes and so just sampling these functions at the current values of and