Analysis of FoG

Foundation model accuracy test: flip overs -> 0, stands normally -> 1.

start of training (1/3)

outputed scores per frame

acc: 73%

mid of training (2/3)

outputed scores per frame

acc: 70%

end of training (3/3)

outputed scores per frame

acc: 96%

Averaged reward over the whole episode during the training. They all increase, indicating the RL optimization part is doing well.

Page updated

Google Sites

Report abuse