SIDM
Maintainer: Sean J. Wang, sjw2@andrew.cmu.edu
Required Dependencies
- Numpy is used for storage of parameters and small math operations
- PyTorch is used for network training
- wheeledRobotSimPybullet is the simulation used to gather data. This package also contains the dataset class used for the replay buffer. It also contains state transformations used during training.
Dataset
Dataset is generated using a PyBullet simulated wheeled robot. Two different systems are provided. The first has a novice and expert robot that are more similar to each other.
Running SIDM Domain Transfer Learning
After the dataset is generated, execute "sidmTraining.py" to run the SIDM training procedure. "sidmNetworks.py" contains all networks needed for SIDM as well as "sidmLosses" class used to calculate all training losses.