Thieso / focops

Pytorch Implementation for First Order Constrained Optimization in Policy Space (FOCOPS).

Implementation for First Order Constraint Optimization in Policy Space (FOCOPS).

Link to paper https://arxiv.org/abs/2002.06506

Requirements

pytorch >= 1.3.1
gym >= 0.15.3
mujoco-py >= 1.50.1.0
For the circle experiments, please also install circle environments at https://github.com/ymzhang01/mujoco-circle.

Implementation

Example: Humanoid task in the robots with speed limits experiments (using the default parameters)

python focops_main.py --env-id='Humanoid-v3' --constraint='velocity'