Option Encoder: A Framework for Discovering a Policy Basis in Reinforcement Learning

Option discovery and skill acquisition frameworks are integral to the functioning of a hierarchically organized Reinforcement learning agent. However, such techniques often yield large number options or skills, which can be represented succinctly by filtering out any redundant information. Such reduction decrease required computation while also improving performance on target task. To compress an array option policies, we attempt find policy basis that accurately captures set all options. In this work, propose Encoder, auto-encoder based framework with intelligently constrained weights, helps discover collection policies. The used as proxy for original skills in suitable framework. We demonstrate efficacy our method grid-worlds evaluating obtained downstream tasks qualitative results Deepmind-lab