What is a policy in reinforcement learning? [closed]

Answer 1

Here is a succinct answer: a policy is the 'thinking' of the agent. It's the mapping of when you are in some state s, which action a should the agent take now? You can think of policies as a lookup table:

state----action----probability/'goodness' of taking the action
  1         1                     0.6
  1         2                     0.4
  2         1                     0.3
  2         2                     0.7

If you are in state 1, you'd (assuming a greedy strategy) pick action 1. If you are in state 2, you'd pick action 2.