Best Way to Condition Agents on Input Images

Hello,

I am sorry for this potentially stupid question, but I was wondering about best practices when condition agents on input images. Ideally, I'd give the agent an image at the beginning of an episode and compare its progress to the image with visual observations. I thought about using the low level python API, but I am not sure if there isn't a more elegent way that doesn't require me to also add custom training code to the agent.

Best and thanks in advance,
Stephanie

 

Why not use a camera sensor ? or the render texture sensor ?

• CameraSensorComponent - Uses images from a Camera as observations.
• RenderTextureSensorComponent - Uses the content of a RenderTexture as observations.

https://github.com/Unity-Technologi...ironment-Design-Agents.md#visual-observations

 

oh thanks so much! I'll try it with RenderTexture observations!

 

you welcome