This repo contains simulation scripts and assets for the ICRA 2021 paper, "Sim-to-real for robotic tactile sensing via physics-based simulation and learned latent projections." (paper | website)
The script provides a simple example of how to import the BioTac assets into NVIDIA Isaac Gym, launch an FEM simulation with multiple indenters 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_biotac.py
- See code for available command line switches
- View
results.hdf5
- File structure is
timestep / feature / environment / data
- 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