Learning Tasks in Intelligent Environments via Inverse Reinforcement Learning

摘要

One of the most common objective for an Intelligent Environment (IE), especially in case of applications of Ambient Assisted Living, is to support a user in everyday tasks. A complex task can be effectively represented as a workflow; i.e., a composition of simpler activities. This research activity aims at developing methodologies and tools able to make a robotic system learns a task from demonstrations in an IE. The large number of accessible sensors in the intelligent environment, as well as the availability of advanced services such as those for activity recognition and situation awareness, can facilitate either the recognition of the action to be executed at the i <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">th</sup> step of the workflow and the verification of the correct execution of the task. In this paper, we present an hybrid approach based on Inverse Reinforcement Learning (IRL) to learn from observations of the behavior of an expert and on forward Reinforcement Learning to correct and improve the learned behavior. We also present the high-level architecture of the IE that supports the learning process.