Abstract
Q-learning, the most popular of reinforcement learning algorithms, has always included an extension to eligibility traces to enable more rapid learning and improved asymptotic performance on non-Markov problems. The λ parameter smoothly shifts on-policy algorithms such as TD(λ) and Sarsa( ) from a pure bootstrapping form ( λ= 0) to a pure Monte Carlo form (λ = 1). In off-policy algorithms, including Q(λ), GQ( ), and off-policy LSTD(λ), the parameter is intended to play the same role, but does not; on every exploratory action these algorithms bootstrap regardless of the value of , and as a result they fail to approximate Monte Carlo learning when λ= 1. It may seem that this is inevitable for any online off-policy algorithm; if updates are made on each step on which the target policy is followed, then how could just the right updates be ‘un-made’ upon deviation from the target policy? In this paper, we introduce a new version of Q(λ) that does exactly that, without significantly increased algorithmic complexity. En route to our new Q(λ), we introduce a new derivation technique based on the forward-view/backward-view analysis familiar from TD(λ) but extended to apply at every time step rather than only at the end of episodes. We apply this technique to derive first a new off-policy version of TD(λ), called PTD(λ), and then our new Q(λ), called PQ(λ).