A.I Learns To Walk (Reinforcement Learning - RockRL)

[@NeededANewName1]

Thank you for all the tutorials, they really help a lot! Do you know if RockRL can be used with CNNs and images for inputs? I tried to implement it but I keep getting an " 'p' must be 1-dimensional" error which I'm pretty sure is because it only takes observations that are 1 dimensions. Unless there is some setting I have overlooked. If not do you know any other library that can do that? Preferably with tensorflow as that is my preferred NN and CNN library. Thank you in advance!

[@PyLessons]

Hey, thanks for support. Yes, RockRL can use images as input and you can use CNNs for that. Maybe you are doing something wrong, because I tested it everything was fine. Open github issue if you want to discuss this further, I'll try to help. Post your code within your question

[@NeededANewName1]

@@PyLessons That is great! If it works I don't have to look for another library. And yeah I properly am doing something wrong in that case. I was a bit hesitant trying to debug it before because I was not even sure it was possible. I will work with it some more before opening a github issue. If you still have your test could you upload it to the github as an example? I think that would really help. Thank you.

[@NeededANewName1]

​@@PyLessons Found the issue, my CNN actually just had a weird shape which caused an error further down. I was using a Dense layer before a Flatten which does not work and the error just happen to be " 'This variable I don't know further down in the library' has to be in 1 dimension" and since I was not sure if RockRL could even do multy dimension input I just assumed the error was in the input being 3D and not the dense layer being 3D. So yeah anyway, thank you again for the quick response, so much easier debugging when you know it 'should' work.