Hi Sanket, thanks for your implementation of GAIL. I plan to run it in AI2THOR for my experiment, so I guess there are changes needed in the code. Could you give some tips on how to do it? My AI2THOR already has a wrapper to interface with openAI gym, so I guess the changes are not that much? (https://github.com/TheMTank/cups-rl).
One more thing I would like to clarify with you. Does the original GAIL applicable if my experts trajectories (sample data) are for the same task but are in different environment? My gut feeling is yes, otherwise can just simply adopt behavioral cloning. But I would like to double check with you.
Hi Sanket, thanks for your implementation of GAIL. I plan to run it in AI2THOR for my experiment, so I guess there are changes needed in the code. Could you give some tips on how to do it? My AI2THOR already has a wrapper to interface with openAI gym, so I guess the changes are not that much? (https://github.com/TheMTank/cups-rl).
One more thing I would like to clarify with you. Does the original GAIL applicable if my experts trajectories (sample data) are for the same task but are in different environment? My gut feeling is yes, otherwise can just simply adopt behavioral cloning. But I would like to double check with you.
Thank you for replying.