the implementation is test on gym games Dependencies: tensorflow r1.3 gym 0.9.2
Proximal policy optimization Algorithms, has some of the benefits of trust region policy optimization(TRPO), but it is much simpler to implement for TRPO is a constained problem and the alogrithm need to use conjugate gradient algorithm to updata the policy's parameter, which need to construct the Fisher information matrix.