Instruction to run the experiments

[linhlpv]

Hi @PaulDaoudi ,

Thank you for your amazing works! I'm trying to run your code and reproduce your experiments in the FOOD's paper. I have looked at readme file but couldn't find any instruction about it. Could you please help me and give some instruction to run the experiments and get FOOD's results.

Thank you so much and have a good day. Best. Linh

[PaulDaoudi]

Hi Linh,

I am glad that you took interest in our paper!

First, please note that the code is not entirely finished: some (minors) errors of my FOOD reimplementation (closely following the original code) must be corrected. It should be finished in the following weeks.

Second, the original trained policies are no longer available, but you can recreate them to gather the target trajectories and obtain the results. Here are the steps to follow:

• Train an RL policy solely in the source and save the associated weights,
• Gather trajectories of the target environment with this policy,
• Filter the trajectories by removing any "bad trajectories" if they exist (for example the ones that lead to the robot's direct fall),
• Train FOOD with the filtered trajectories.

I will keep you updated when the code will be entirely finished.

Thanks again for your patience and interest in our work!

Best,

Paul

[linhlpv]

@PaulDaoudi ,

Thank you for your reply.

As you mentioned about the original code, may I ask you where I can find it? Second, about training the policy to gather the source trajectories, how can I train the policy? Do you provide any source code for training and gathering (filtering bad trajectories too) in your repo?

Thank you again and have a nice day. Best, Linh