python train2.py --model deform_contactnet_pointnet2 --use_normals --log_dir deform_contactnet_pointnet2 --batch_size=8 --epoch=10 --use_wandb --gpu=1
python train.py --model deform_contactnet_pointnet --use_normals --log_dir deform_contactnet_pointnet --batch_size=9 --use_wandb --epoch=10 --gpu=0
python planning.py --object=bowl_ycb --use_vis
The script provides a simple example of how to import the Sponge assets into NVIDIA Isaac Gym, launch an FEM simulation with multiple objects across multiple parallel environments, and extract useful features (net forces, nodal coordinates, and element-wise stresses).
- Clone repo
- Download NVIDIA Isaac Gym
- Follow provided instructions to create and activate
rlgpu
Conda environment for Isaac Gym
- Follow provided instructions to create and activate
- Install
h5py
package via Conda
- Execute
sim_sponge.py --object="target_object" --num_envs=1 --youngs=1000
- See code for available command line switches
- View
results/object/object_youngs
- File structure is
action_success / contact_indexes / gripper_ori / normal_forces_on_nodes / press_locations / press_forces /sponge_position_at_force
- File structure is
- Error:
cannot open shared object file
- Add
/home/username/anaconda3/envs/rlgpu/lib
toLD_LIBRARY_PATH
- Add
- Warning:
Degenerate or inverted tet
- Safely ignore
- For questions related to NVIDIA Isaac Gym, see official forum