cbfinn / gps

Guided Policy Search
http://rll.berkeley.edu/gps/
Other
597 stars 239 forks source link

CostAction first order terms #26

Closed dementrock closed 8 years ago

dementrock commented 8 years ago

In https://github.com/cbfinn/gps/blob/master/python/gps/algorithm/cost/cost_action.py#L28, the term dc/du is computed as (line 28):

lu = self._hyperparams['wu'] * sample_u

Should there be a negative sign in front, since this should compute the expansion of 0.5*(u-uref)^T*R*(u-uref)?