Work in Progress - Enhancing Reinforcement Learning Class Curriculum using a Matlab Interface Library for use with the Sony AIBO Robot

摘要

In an effort to mitigate the inherent complexities of embedded robotics programming, a novel Matlab-based interface library with matching firmware is presented. The platform was developed at the University of Tennessee for use in a course on reinforcement learning. Using this software interface, students were able to easily implement machine learning schemes and apply them to the Sony AIBO robot, a flexible platform with diverse capabilities. The behavior development process for the AIBO is traditionally limited to programming in a C++ environment. To avoid the overhead of such tedious programming, the Matlab interface was designed to allow the user to both implement functions and behaviors already developed for the AIBO, as well as to construct new behaviors using existing components as building blocks. Moreover, since the interface relies on a bidirectional network socket connection, the user is offered full control of the robot from any (remote) networked computer.