Transfer Learning across heterogeneous robots

[haitham973]

I have a small question concerning the cool results that you have attained:=) Given that the target robot operates within the same environment it is a bit easier to do the transfer among different task. How did you do the action mappings between the 2 robots given that they have different action spaces ? Was it manually encoded ? As the last part of the video says that the "actions are not mapped and imitated", then how did you utilize and transfered the policy to the target robot ?

Thanks :=)