Deep reinforcement Learning

Project: The goal of this course was to achieve impressive performance on the games from the OpenAI gym game library: Acrobot, mountain Car and cart pole. In combination with solving offline problems and multi robot problems.

Solution: To achieve this, a multitude of concepts were applied and the efficiency between these concepts were compared. The high-level concepts that were implemented were: Stabilization of approximated value estimation, Modern actor-critic methods, Planning as inference, deep networks and various deep exploration techniques.