Deep Q-Network with Predictive State Models in Partially Observable Domains

Table 1

The best reward of three methods.

	CartPole-v1	Swimmer-v1	Reacher-v1
DRQN	200	56	−1.15
DQN-1frame	54	40.58	−6.43
RPSR-DQN	200	59.52	−0.02
RPSP	158	38.96	−57.78