Reinforcement Learning Toolbox: Episode Q0 does not change ., DDPG agent

Hi :-)

I am training my ddpg agent with matlab.

I saw something weird with my trainig graph at reinforcement learning episode manager.

this is my plot, and as you can see,, Q0 value never follows or go near the average reward.

Does this mean that there is something wrong with my critic network? but I can see the traing procedure is working quite properly I guess,,,

Please help!!

[Mokhtar]

Hello, i am having the same Problem, did u find any solutions ?

Answer 1

[Berk Agin]

Hello,

I also see that this problem on my training data. I couldn't understand and wonder the solution. Have a nice day.

[Andreas Götz]

Same with my training data.....