To prove that A is NP-hard, you have to do the opposite. Show how you transform every instance of B into an instance of A, so that a solution of A implies a solution of B. Then if you had a solver for A, you could use it to solve B, which is impossible unless P=NP (since B is known to be NP-hard). In this case, you have to transform TSP to Pac-Man, not the other way around. Some technicalities aside (which are important, nevertheless), this is how hardness proofs go, sorry if it was a bit obscure.
Now it's clearer why he used a form of Pacman where pacman could only visit each node once, and why it's not as trivial as i made it seem.
Input: A, B : integers
Output: directed graph G(V, E, w) (V = vertices, E = edges, w = weight fn)
V = {c1, c2, c3}
E = {(c1, c2), (c2, c3), (c3, c1)}
w(c1, c2) = A
w(c2, c3) = B
w(c3, c1) = 0
shortest tour is the only valid tour whose length is clearly A + B.