I can use Generative Adversarial Imitation Learning(GAIL) to solve CartPole-v0 but fail in MountainCar-v0. The original GAIL proposed by Jonathan Ho, et al. used trust regio
