1. Using Reinforcement Learning to Learn Self-Motivation

Advisor: Dr. Doug Blank

Senior Project 2009

1.1. Proposal

1.2. Brief Description

I am investigating whether a robot can track a moving object (in this case another robot) and predict its next move using reinforcement learning.

1.3. Updates

Jan 29, 2009

What I have done till now:

• expand on experiment done by Marshall et al. (2004)

• wrote proposal

Next steps:

• simulation or robots?
• simulation
1. put up a wall: what happens when target robot goes behind a wall? can the developing robot correctly figure out when he'll come out the other side? or where is he behind the wall at any given moment?
2. add movement to the developing robot? give it more functionality?
3. can it track something else?

iv. use a different model?

• what if the developing robot doesn't pay attention to the new robot?

• robot
1. what will i use to track? camera? IR?

• write background/related work

Feb 3, 2009

Robots! - Will use Scribbler to reproduce experiment. (Change from initial proposal)

After doing simple environment, perhaps add some complexity - wall?

Next:

• Play around with Scribbler

• Become familiar with image manipulation

• Study model algorithm used in Marshall et al.

Feb 5, 2009

Learn to talk about it! (See Brief Description)

1.4. Relevant Readings

An Emergent Framework for Self-Motivation in Developmental Robotics (Marshall et al., 2004)

The Introspective Robot: Using Self-Prediction to Improve Robot Learning (Marshall et al., 2008)

The Multiple Roles of Anticipation in Developmental Robotics (Blank et al., 2005)

Bringing up robot: Fundamental mechanisms for creating a self-motivated, self-organizing architecture (Blank et al.)

Chap 7 of What is Thought by Eric Baum

Reinforcement Learning by Richard S. Sutton and Andrew G. Barto