As a student of computer science very first thing I learned in my very first year, was art of programming using object oriented approach. Key to this paradigm of programming was “Abstraction”.
“Hiding details that don’t matter at certain level of system function.” one of my book read. Easy! But not trivial.
Abstraction makes human decision making more organised and efficient in terms of resources and efforts. Achieving similar organisation, understanding and its use for planning and transfer has been long standing challenge in Artificial intelligence and Robotics.
This article talks about possibility of use of temporal abstraction in robot planning, so let’s understand what temporal abstraction means from a robotic point of view.
Imagine an arm trying to set up a table. This task can involve decision making at multiple level starting from high-level decisions like plate goes in centre and fork and spoon goes to either side. Level below it can involve control task of actually grabbing stuff and so on to the lowest level of abstraction which involves how much voltage is actually passed to actuators in order to perform a certain movement.
I will be discussing some of the recent research work in Reinforcement Learning which can be used for planning at multiple level of this temporal abstraction.
In each post on this page one paper is discussed. (Link to the paper and author information at the end of the post.)
“I am not claiming any of this work as mine this is just a way to put them together for better understanding and reference”