Episode Q0 increases exponentially

Can anyone explain why episode Q0 in RL increases exponentially after convergence of reward to a suboptimal policy?

Answer 1

[Emmanouil Tzorakoleftherakis]

Hello,

Please take a look at this answer for some suggestions. Normalizing observations, rewards, and actions can also help avoid situations like these.

Hope this helps

[DAMODARAN B.K]

is episode Q0, criticnetwork output or target value?