Supervised Reinforcement Learning - Application to an Embodied Mobile Robot

ISBN-13: 9783836428064 / Angielski / Miękka / 2007 / 112 str.

Karla Conn

Opis

Can machines be taught? If so, what methods are useful for teachingmachines? Machine learning is a field focused on systems that can learnthrough their own experiences and evaluation. Programmers could encode allbehaviors for a task, but this process quickly becomes limited to condensedproblems. Therefore, scientists have turned to methods with adaptability, takingcues from biological systems (including the human brain) to solve morecomplex problems in varied environments. This book describes two experimentsimplementing supervised reinforcement learning on a real, mobile robot.One tests the robots reliability in completing a navigation task it hasbeen taught by a supervisor. The other, in which obstacles are placed alongthe path to the goal, measures the robots robustness to changes in environment.Experimental analysis answered: How quickly can the robot find thegoal? How much reward does the robot amass? How often does the robotfail in the task? How closely does the robot match the supervisors actions?This book is addressed to those looking for means to teach robots aboutrewards/punishments, such as researchers in Robotics, Machine Learning, andEngineering.