Poor results

[robertvo]

I trained the agent overnight on fast computer, resulted in these stats:
experience replay size: 30000
exploration epsilon: 0
age: 3662244
average Q-learning loss: 0.0158955232113
smooth-ish reward: 0.870696559286462

The result is the bot is spinning in place in the wall corner.
This is not working as it should.
Not sure if it's the bot of the DQN itself :(