Phasic Policy Gradient

                Karl Cobbe 1 Jacob Hilton 1 Oleg Klimov 1 John Schulman 1

               Abstract              can be used to better optimize the other.
    We introduce Phasic Policy Gradient (PPG), a re-     However, there are also disadvantages to sharing network
    inforcement learning framework which modifies       parameters. First, it is not clear how to appropriately balance
    traditional on-policy actor-critic methods by sepa-    the competing objectives of t ...