learnRL

Learning reinforcement learning (in MATLAB)

This software a playground and is aimed specifically at studying reinforcement learning (RL) in detail with a rich variety of settings. The core of the playground is based upon a model of a mobile robot, referred to as the so called "extended non-holonomic double integrator" (ENDI). See these notes for its description. A flowchart of the overall code can be found in here. Basically, an agent (referred to also as the "controller") is attached to the environment (the system) and generates actions so as to minimize running costs (also called rewards or stage costs) over an infinite horizon in future. The specific objective in this software package it so park the robot. The controller is multi-modal and allows comparison with various baselines (nominal parking controller, model-predictive controller with and without on-the-fly model estimation).