Rllib has amazing features for reinforcement learning, but unfortunately, I couldn’t find a way to input images as an observation without flattening them (I basically want to use convolutional neural network). Is there any way to input image observations in models using rllib library? I basically want to use convolutional neural network
Hi @dato_nefaridze ,
It is the common case in RLlib to use convolutional networks for all the Atari Learning Environment settings we have tuned. For example, if you select an Atari environment that returns 2D observations, RLlib will build a standard model with convolutional layers for you.
Please post more info about how you are trying to achieve what you want at the moment. Are you trying to provide a custom model but the inputs are flat?
Cheers
@arturn Thanks for reply: I am trying to implement algorithm for real drone which gets observations from the camera, so is there any way i can do it?
Yes, there is. If you want to implement it, I recommend reading through the user guides. They will do a much better job of explaining most steps than I can do here.
@arturn : could you provide an example how to take a 2d image at input?