Chapter 3 SARSA

Page 60, Section 3.2.1 Intuition for Temporal Difference Learning

Page 60, the first square bullet:

(s0,aUP): The agent moves out of the corridor, receives a reward of 0, and theepisode terminates, so Q(s0,aUP)=0.(s_0, a_{UP}): \text{ The agent moves out of the corridor, receives a reward of 0, and the} \newline \text{episode terminates, so } Q^*(s_0, a_{UP}) = 0.

contains a typo in the index of the state. It should read:

(s1,aUP): The agent moves out of the corridor, receives a reward of 0, and theepisode terminates, so Q(s1,aUP)=0.(s_1, a_{UP}): \text{ The agent moves out of the corridor, receives a reward of 0, and the} \newline \text{episode terminates, so } Q^*(s_1, a_{UP}) = 0.

Page 62, Section 3.2.1 Intuition for Temporal Difference Learning

Page 62, Figure 3.3, Episode 4, Time step 5, the target value calculation:

0+0.9+0.9=0.810 + 0.9 + 0.9 = 0.81

contains a typo, the second + should have been x. It should read:

0+0.9×0.9=0.810 + 0.9 \times 0.9 =0.81

Page 67, Section 3.4 SARSA Algorithm, Algorithm 3.1

Algorithm 3.1 SARSA, line 13:

θ=θαθJ(θ)θ = θ − α∇θJ(θ)

Contains a typo; the loss function J should have been L:

θ=θαθL(θ)θ = θ − α∇θL(θ)

Last updated

Was this helpful?