6fd6b030c6afec018415662d0db43f9d-AuthorFeedback.pdf
–Neural Information Processing Systems
Half Cheetah is indeed a somewhat limited benchmark. We chose it because it trains31 quickly, and the base algorithm required no additional tuning to work out of the box (unlike other environments,32 where we found existing implementations would perform unreliably or require very large amounts of training).33
Neural Information Processing Systems
Feb-12-2026, 13:28:21 GMT