Reinforcement learning : How to define custom environment with multiple image based observations

Hi,

I am trying to create a custom agent with image based observations. How can assign an image to rlNumericSpec.

I have created ObservationInfo as follows since the image is 320X180. Normally the image is RGB even if I converted to gray I do not know how to assign in to ObservationInfo in the customer reset and step functions.

ObservationInfo = rlNumericSpec([320 180]);

Even if I solve this I require RGB images and later on more than one images.

Any ideas?

Best regards

Answer 1

[Emmanouil Tzorakoleftherakis]

For grayscale images, take a look at this example. For rgb, maybe the following would work

ObservationInfo = rlNumericSpec([320 180 3]);

[Dr. Ali Peker]

Thank you I have converted the image using im2double in custom reset and step functions. for RGB I have used [180 320 3] it works.

[DAMODARAN B.K]

from visualiser plot how did you take images?