Computer Research Association Distributed Mentoring Project CRA - DMP
@ The University of Oklahoma, Summer 2006

Marina Djerekarova

• University: Florida International University
  
• Major: Computer Science
  
• Minor: Physics
• Grade Level: Junior
  
• Nationality: Bulgarian
  I plan to graduate in Summer 2007.
  At Florida International University I am involved in the Tennis Club and the Society of Physics Students.
• Email: mdjer001@fiu.edu
  

MY SUMMER PROJECT

Reinforcement Learning on the Bions

Bions in the Lab	A bion, AVR programmer and a circuit. The circuit allows printf output to be displayed through a connection with a program called kermit.	A close look of an individual bion.

This summer I am participating in the DMP Program at the Stephenson Research Center, which should be part of Oklahoma University, but it is definitely not a walking distance away. My mentor is Dr. Amy McGovern.

I am working on the bions and my project is to implement Reinforcement Learning on them. I am using Windows Virtual machine through Red Hat Linux OS and C programming language. The Bions have to be programmed in AVR Studio 4 IDE and that software can only be installed in Windows. So life gets complicated sometimes with so many open applications, 2 monitors and the bion hardware. But it's new and interesting. And I am getting more experience using Linux.

The Bion Hardware implementation is:
• Atmel Mega8 Microcontroller
• 1 K memory, 8 K instruction space
• electric speaker
• 4 bright, blue LEDs

The Project constitutes in the following:
• Get some idea of what Reinforcement Learning is.
• Understand the documentation and the code which is already written for the bions.
• Program a function, which will allow the bion to play short and simple songs.
• Program a few different short tunes into a given amount of bions (the bions which will be used for learning).
• Reinforcement Learning (RL) on the Bions Implementation
  • Agent - individual bion.
  • Action Representation -possible notes to play.
    From each possible state, the bion could play only a limited number of nearby notes, so that the learning can be faster.
  • State representation
    • proximity or not
    • interaction with other people in 4 different directions
    • interaction with other bions from 4 different directions
  • Award Representation
    When the person approaches the bions they will start playing a song. If the person likes it , he/she will stay in proximity and a reward will be given if the person stays for the entire song. Otherwise no reward will be given.

  back to top