Skip to main content
1 of 5
Neil G
  • 15.5k
  • 3
  • 49
  • 91

First of all, there's no reason that an agent has to do the greedy action; Agents can explore or they can follow options. This is not what separates on-policy from off-policy learning.

The reason that Q-learning is off-policy is that it updates its Q-values using the Q-value of the next state $s'$ and the greedy action $a'$. In other words, it estimates the return (total discounted future reward) for state-action pairs assuming a greedy policy.

The reason that SARSA is on-policy is that it updates its Q-values using the Q-value of the next state $s'$ and the current policy's action $a''$. It estimates the return for state-action pairs assuming the current policy.

The distinction disappears if the current policy is a greedy policy. Such a learner would be both off-policy and on-policy. However, such an agent would not be good since it never explores.

Have you looked at the book available for free online? Sutton, Richard S., and Andrew G. Barto. Reinforcement learning: An introduction. Vol. 1. No. 1. Cambridge: MIT press, 1998.

Neil G
  • 15.5k
  • 3
  • 49
  • 91