Learning Deformable Object Model with GNS

Codes for learning deformable object model with GNS

Install

• create conda environment: conda create -n deformable_gns python=3.7 pip
• run pip install pybullet for pybullet
• Install requirements files according to https://github.com/deepmind/deepmind-research/tree/master/learning_to_simulate (requires Python3.7-): pip install -r learning_to_simulate/requirements.txt
• Downgrade Protobuf to avoid potential problems of tensorflow: pip install --upgrade "protobuf<=3.20.1"
• Install sklearn: pip install -U scikit-learn
• Install IPOPT: sudo apt-get install gcc g++ gfortran git patch wget pkg-config liblapack-dev libmetis-dev
• Install IPOPT python wrapper: conda install -c conda-forge cyipopt (required Python 3.6+)
• Install opencv: conda install -c conda-forge opencv

Usage

• You can use chmod +x run.sh & run.sh for collecting the data and train the network
• For details, run python collect_rope_data_2d.py for collecting rope data in tfrecord format.
• You can use python learning_to_simulate/show_data_message.py to test whether the data is recorded correctly.
• run python learning_to_simulate/evaluate.py for generating rollouts
• run python learning_to_simulate/render_rope.py for the results
• run python planning.py for planning example (a pre-trained model was put in the learning_to_simulate/models)
• for speeding up the planning, you can play with the optimizaiton options tol and max_iter

Troubleshooting

Contact changhaowang@berkeley.edu