Learning behavior styles with inverse reinforcement learning

by Jim Hill on January 23, 2011

in Assignment 1

1. Sentence: We present a method for inferring the behavior styles of character controllers from a small set of examples (1st sentence, 1st paragraph)

2. Problem: Developing reward functions for the creation of character controllers is difficult and imprecise to do manually. Instead, build a system that will generate the reward function given several examples.

3. Key Idea: Use Inverse Reinforcement Learning to generate a character controller using several examples.

4.a What the paper does: Provides a method for automatically generating a character controller which displays the desired behavior style while attempting to reach a task goal. These character controllers can then be blended and used to achieve different task goals.

4.b What it could be used for: Developing general purpose character controllers which will achieve task goals in a behaviorally satisfying way.

5. Resources: video (http://grail.cs.washington.edu/projects/learning-behavior-styles/learning%20behavior%20styles.mov) on author’s web page.

Previous post:

Next post: