I'm familiar with DQN libraries and I think I should be able to build the model and provide the code to train it.
However, the problem is that this technique is somewhat hard to code it right, and get it work right, thus I can not guarantee a success at this point, without trying it out.
My suggestion is that, provide me a way to run the "game" / simulation part, and I will see if I can get the DQN work.
I'll let you know once I could get it work and show you the results. If you are happy with the results, you may proceed to the actual payment.
By the way, just from quriosity, is there any particular reason for using reinforcement learning over PID / Model based control?