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

Figure 2

The control tasks in our experiments: (a) CartPole-v1, (b) Swimmer-v1, and (c) Reacher-v1.

(a)

(b)

(c)