With the current version of Acrobot-v1, an angle could become arbitrarily negative, eventually leading to nans. This is due to a bug in the wrap function of acrobot.py. For example, wrap(-2, -1, 1) gave Array(-2., dtype=float32, weak_type=True), which was probably not intended.
I suppose that the error wasn't detected by the tests, but I wasn't sure how to fix that.
With the current version of
Acrobot-v1
, an angle could become arbitrarily negative, eventually leading tonan
s. This is due to a bug in thewrap
function ofacrobot.py
. For example,wrap(-2, -1, 1)
gaveArray(-2., dtype=float32, weak_type=True)
, which was probably not intended. I suppose that the error wasn't detected by the tests, but I wasn't sure how to fix that.