action-conditional-rkn

Pytorch code for CoRL 2020 paper Action-Conditional Recurrent Kalman Networks For Forward and Inverse Dynamics Learning

Dependencies

• torch==1.3.1
• python 3.*

How to Train

Forward Dynamics - With action-conditional-rkn as the working directory execute the python script python experiments/PamForward/pneumaticArm.py

Inverse Dynamics - With action-conditional-rkn as the working directory execute the python script python experiments/Franka/frankaArm.py

Datasets

Forward dynamics learning uses subset of muscular robot table tennis data from here. Inverse dynamics learning uses inhouse data collected from Franka Emika Panda