grasp_diffusion

grasp_diffusion copied to clipboard

improve batch-wise grasp generation in sampling scripts

1

To resolve issue #4 - added consistent use of `batch_size` and `device` variables - added sampling loop to support desired n_grasps - added `device` and `batch_size` as cli args -...

kuldeepbrd1

Minor: `n_grasps`, `n_envs` and `batch` usage in sampling scripts

1

The scripts in [`scripts/sample`](scripts/sample) do not use n_grasps from the cli arguments. `n_grasps` and `n_envs` in not used. Instead it always generate number of grasps equal to batch size specified...

kuldeepbrd1

Paper - Algorithm 4: Trajectory optimization pipeline

3

Hi @TheCamusean ! Thank you for sharing your awesome work!! Do you have an example of Algorithm 4 in the paper with a full robot arm? Thank you, Juan

jucamohedano

Good morning: I'm folllowing proposed steps to generate the dataset starting from ShapeNetSem. When running grasp_difussion/scripts/create_data/simplify.sh changing the paths, an error arose: ``` 100f39dce7690f59efb94709f30ce0d2.obj manifold 100f39dce7690f59efb94709f30ce0d2.obj temp.watertight.obj 20000 simplify: /Manifold/3rd/igl/per_vertex_point_to_plane_quadrics.cpp:83:...

ignpaub

some question about loss function

1

Hello author @TheCamusean This is really a great job, but I have some questions about the loss function. What is the difference between ProjectedSE3DenoisingLoss and SE3DenoisingLoss. The loss mentioned in...

o0stinger0o

m0dd0

Hi, I was wondering if the grasp_dif_multi model is trained on the entire ACRONYM dataset or a subset of that dataset? In case it is a subset, do you have...

shukla-yash

Loss not decreasing

3

Is it expected to see the loss not go down during training? Loss kind of stagnates after the first epoch which does not make any sense. What am I missing?

SrinidhiBharadwaj

Hi, the google drive link for the prepared data seems to be failing. Can you help renew the link. thanks!

cyugai

The paper mentions the use of signed distance for each object. However, the preprocessing code makes use of unsigned distance. Is this correct? (`absolute=True` is used) https://github.com/robotgradient/grasp_diffusion/blob/59b633c5bdfec1c9c5505ea0f4f8d963beeed252/scripts/create_data/generate_mesh_sdf.py#L43-L58

stalhabukhari