Learning Football Evaluation for a Walking Robot
Abstract
The Ambler is a six-legged robot designed to walk in irregular and rugged terrain. Footfall evaluation is the problem of predicting the goodness of footfall locations given the current status of the Ambler and properties of the terrain. We use an inductive learning technique that implicitly correlates terrain features to footfall stability and traction. The learning method also possesses the desirable characteristics of adaptability to different environments, noise tolerance, ability to be trained using relative measures, efficiency and extensibility. This paper describes how a feature-based neural net and a user-specified cost classification scheme can be used to do the evaluation and training. The approach sketched here is applicable to cases where a real-valued function is to be learned from relative measures.
Cite
Text
Hsu and Simmons. "Learning Football Evaluation for a Walking Robot." International Conference on Machine Learning, 1991. doi:10.1016/B978-1-55860-200-7.50063-5Markdown
[Hsu and Simmons. "Learning Football Evaluation for a Walking Robot." International Conference on Machine Learning, 1991.](https://mlanthology.org/icml/1991/hsu1991icml-learning/) doi:10.1016/B978-1-55860-200-7.50063-5BibTeX
@inproceedings{hsu1991icml-learning,
title = {{Learning Football Evaluation for a Walking Robot}},
author = {Hsu, Goang-Tay and Simmons, Reid G.},
booktitle = {International Conference on Machine Learning},
year = {1991},
pages = {303-307},
doi = {10.1016/B978-1-55860-200-7.50063-5},
url = {https://mlanthology.org/icml/1991/hsu1991icml-learning/}
}