Voronoi State Partitioning for Linear Reinforcement Learning Policies

Voronoi State Partitioning for Linear Reinforcement Learning Policies ■

Senne Deproost, Ann Nowe

Abstract ■

We want to mimic a DRL policy using a set of smaller linear models to increase interpretability. To accomplish this, we learn to partiton the state space into Voronoi cells based on the capability of those models to operate in such region