Identical to #17, except since we're passing in the last 4 frames with FrameStack now we can feed the neural net (frame(-3) - frame(-2), frame(-2) - frame(-1), frame (-1) - frame(0), frame(0)), and see if it does better. this simulates manual feature encoding for dynamics.
vlad17
commented
6 years ago
Identical to #17, except since we're passing in the last 4 frames with
FrameStacknow we can feed the neural net(frame(-3) - frame(-2), frame(-2) - frame(-1), frame (-1) - frame(0), frame(0)), and see if it does better. this simulates manual feature encoding for dynamics.