Producing agents that are effective in video game and video game-like environments is a diffcult task for which powerful solutions are needed. One such solution, shaping, has been applied to standard evolutionary algorithms in the past in environments such as the NERO video game, and had considerable success. Another solution, coevolution has been shown to be a promising approach, but has been diffcult to control. In this paper, the shaping approach is applied to coevolution in an attempt to enhance the already powerful effects of coevolution in competitive environments. Several example automated shaping methods are introduced and tested in a simple capture-the-flag-like environment using the OpenNERO research platform and rtNEAT for neuroevolution. This paper demonstrates that each of these shaping methods is superior to a control, and further analyzes the results in order to determine general rules for shaping coevolution. While different methods of shaping apply differently to different environments, using the approach detailed here, it should be possible to use coevolution to create intelligent agents in a variety of situations.

PDF

Technical Report HR-11-01, Department of Computer Science, The University of Texas at Austin, 2011.

@techreport{dziuk:ugthesis11, title={Creating Intelligent Agents through Shaping of Coevolution}, author={Adam Dziuk}, number={HR-11-01}, institution={Department of Computer Science, The University of Texas at Austin}, type={Undergraduate Honors Thesis}, url="http://nn.cs.utexas.edu/?dziuk:ugthesis11", year={2011} }