Missing demonstration data for example_birp.py

[setowenGit]

Hello, I’d like to express my appreciation for your work on the BiRP paper. It has been an insightful read, and I’m excited to dive deeper into the implementation.
I’ve been trying to follow along with the example_birp.py to better understand the concepts, but it seems to be missing the necessary demonstration data, which is causing the script to fail. Could you kindly provide the required data or let me know where I can find it?
Thank you so much for your time and assistance. I look forward to your response!

[Skylark0924]

Sorry for that. Since the data folder is quite large, I put the data on Google drive and forgot to mention it in README. Here is the link: https://drive.google.com/uc?id=1pOzD61CQJcy4L2hXveT1cGiD0AkIDt_c&export=download

[Skylark0924]

This link was mentioned in the install script

Rofunc/scripts/install_w_isaacgym.sh

Line 30 in 67083ec

gdown https://drive.google.com/uc?id=1pOzD61CQJcy4L2hXveT1cGiD0AkIDt_c&export=download

[setowenGit]

Thanks your reply and data!

[setowenGit]

Thank you for your reply! However, I’m sorry, I still couldn’t find the demonstration data used in example_birp.py. It seems that demo_exp_7.csv is used in the path /home/ubuntu/Data/optitrack_record/2023_03_29, but the data you provided doesn't seem to have the corresponding files.

[Skylark0924]

You can try https://github.com/Skylark0924/Rofunc/blob/main/examples/learning_ml/example_tpgmmbi_RPCtrl.py and https://github.com/Skylark0924/Rofunc/blob/main/examples/learning_ml/example_tpgmmbi_RPRepr.py first. Cause these are the examples shown in the paper to reveal the difference between method variants. The code you mentioned should be the one we used during real experiments, the exp data was not stored and I need some time to find out.

[setowenGit]

Thank you very much for your response. However, I’m afraid I still have some additional questions. When I tried running the example_tpgmmbi_RPRepr.py script, I encountered an error:

I resolved the issue by modifying the code as follows:
demos_dx, demos_A, demos_b, demos_A_xdx, demos_b_xdx, demos_xdx, demos_xdx_f, demos_xdx_augm = repr.get_related_matrix()

This fix worked, but I encountered a new error:

I modified the code to:
follower_traj = reprs[follower]._reproduce(models[follower], follower_prod, ref_demo_idx, self.task_params[follower]['frame_origins'][0][0])

After this, the code ran successfully and I was able to generate the "Reproductions for new situations" figure.

However, when comparing the result with the figure in your BIRP paper, I noticed a significant difference: the meeting point of the two robot arms in my output is clearly not meeting, with a distance between them. I also tried increasing the number of iterations, but this did not help.

I would like to know if my modifications caused the problem or if the code itself still has some potential issues that led to this poor result. I hope that my feedback can help improve this excellent project.

Thank you again for your time and help!

[Skylark0924]

Let me check

[Skylark0924]

I just tried the example code https://github.com/Skylark0924/Rofunc/blob/main/examples/learning_ml/example_tpgmmbi_RPCtrl.py

The coordinate feature I set is to let the bimanual trajectory meet at one point. So I gave demos like these

The original TPGMM, which does not consider coordination, is like this

After considering relative parameterization in control level, we got this