xingyu-lin / vcd Goto Github PK

Thank you for open-sourcing your work. In the ClothDataset._find_and_update_picked_point method, it seems that picked_particles has never been updated:

 #line 284
       if robot_exp:
            new_picker_pos = None
            num_picker = 2
            for i in range(num_picker):
                if pick_flag[i]:
                    if picked_particles[i] != -1:
                        old_pos = vox_pc[picked_particles[i]]
                        new_pos = vox_pc[picked_particles[i]] + action[i, :3]
                        new_vel = (new_pos - old_pos) / (self.dt*self.args.pred_time_interval)

                        tmp_vel_history = (velocity_his[picked_particles[i]][:-3]).copy()
                        velocity_his[picked_particles[i], 3:] = tmp_vel_history
                        velocity_his[picked_particles[i], :3] = new_vel
                        vox_pc[picked_particles[i]] = new_pos

                        picked_velocity.append(velocity_his[picked_particles[i]])
                        picked_pos.append(new_pos)
                else:
                    picked_particles[i] = -1

Is this a bug?

As mentioned in #2,
I downloaded best_state.json from google drive and place them as shown below.
(Since the two files had the same name, I created directories named "edge" and "dyn".)

VCD
|--VCD
|--chester
|--pretrained
 |--dyn
  |--best_state.json
  |--vsbl_dyn_150.pth
 |--edge
  |--best_state.json
  |--vsbl_edge_best.pth
 ︙

Then I ran the following command
python VCD/main_plan.py --edge_model_path=./pretrained/edge/vsbl_edge_best.pth --partial_dyn_path=./pretrained/dyn/vsbl_dyn_150.pth
And I got

Traceback (most recent call last):
  File "VCD/main_plan.py", line 386, in <module>
    main(get_default_args())
  File "VCD/main_plan.py", line 192, in main
    vcd_edge = load_edge_model(args.edge_model_path, env)
  File "VCD/main_plan.py", line 132, in load_edge_model
    vcd_edge = VCConnection(edge_model_args, env=env)
  File "/home/pc/user/VCD/VCD/vc_edge.py", line 33, in __init__
    self.load_model(self.args.load_optim)
  File "/home/pc/user/VCD/VCD/vc_edge.py", line 271, in load_model
    self.model.load_model(self.args.edge_model_path, load_optim=load_optim, optim=self.optim)
  File "/home/pc/user/VCD/VCD/models.py", line 311, in load_model
    ckpt = torch.load(model_path)
  File "/home/pc/anaconda3/envs/softgym/lib/python3.6/site-packages/torch/serialization.py", line 601, in load
    if _is_torchscript_zip(opened_zipfile):
  File "/home/pc/anaconda3/envs/softgym/lib/python3.6/site-packages/torch/serialization.py", line 890, in _is_torchscript_zip
    return 'constants.pkl' in zip_file.get_all_records()
RuntimeError: [enforce fail at inline_container.cc:230] . file in archive is not in a subdirectory archive/: best_state.json

How can I fix this error? Thank you:)

Currently, I'am training the dynamics model with the generated data I made.
but the training seems too slow. it takes 10~20 miniutes for 1 dataloader iteration (1 batch).
Is it normal operation?? I want to know how much it takes for training per 1 epoch or 1 batch in normal case.
I'am using one RTX 2080 Ti gpu and script settings are the default options except for --num_workers=4 because of my pc spec.

I've tried VCD planning with a pre-trained model, but I got an error.

Traceback (most recent call last):
  File "VCD/main_plan.py", line 383, in <module>
    main(get_default_args())
  File "VCD/main_plan.py", line 189, in main
    vcd_edge = load_edge_model(args.edge_model_path, env)
  File "VCD/main_plan.py", line 125, in load_edge_model
    edge_model_vv = json.load(open(osp.join(edge_model_dir, 'best_state.json')))
FileNotFoundError: [Errno 2] No such file or directory: './pretrained/best_state.json'

Where can I find the best_state.json? Thank you.

I am currently exploring research in robotics and came across your project. I am particularly interested in the real robotic arm experiments you have conducted. Would it be possible for you to share the code for these experiments? I believe it would greatly benefit my understanding and further development in this area. Thank you very much for considering my request.

Hi, I would like to install and run your project. But there are complaints about the conda environments. It seems like pyg only supports >=python 3.7 while python-PCL only supports < python 3.7. Can you provide a feasible solution for installing your packages, please?

I fixed the bug while using num_workers = 0, but you have data types unmatch problem between the input data and the weights of the model in the encoders.