Our code is built on top of SAVi codebase. We provide four important python files which cover the algorithm of our proposed AVLEN approach. 

ddppo_trainer.py: Base code to take rollout steps and update policy based on collected experience.  
ppo_trainer.py: Contains rollout collection function and the model evaluation code.
policy.py: Defines model architecture
ppo.py: Python script for updationg model using proximal policy optimization.