REALWORLD ROBOT NAVIGATION BY TWO DIMENSIONAL EVALUATION REINFORCEMENT LEARNING

Hiroyuki Okada

2004

Abstract

The trade-off of exploration and exploitation is present for a learnig method based on the trial and error such as reinforcement learning. We have proposed a reinforcement learning algorism using reward and punishment as repulsive evaluation(2D-RL). In the algorithm, an appropriate balance between exploration and exploitation can be attained by using interest and utility. In this paper, we applied the 2D-RL to a navigation learning task of mobile robot, and the robot found a better path in real world by 2D-RL than by traditional actor-critic model.

Paper Citation

in Harvard Style

Okada H. (2004). REALWORLD ROBOT NAVIGATION BY TWO DIMENSIONAL EVALUATION REINFORCEMENT LEARNING . In Proceedings of the First International Conference on Informatics in Control, Automation and Robotics - Volume 2: ICINCO, ISBN 972-8865-12-0, pages 249-255. DOI: 10.5220/0001136602490255

in Bibtex Style

@conference{icinco04,
author={Hiroyuki Okada},
title={REALWORLD ROBOT NAVIGATION BY TWO DIMENSIONAL EVALUATION REINFORCEMENT LEARNING},
booktitle={Proceedings of the First International Conference on Informatics in Control, Automation and Robotics - Volume 2: ICINCO,},
year={2004},
pages={249-255},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0001136602490255},
isbn={972-8865-12-0},
}

in EndNote Style

TY - CONF
JO - Proceedings of the First International Conference on Informatics in Control, Automation and Robotics - Volume 2: ICINCO,
TI - REALWORLD ROBOT NAVIGATION BY TWO DIMENSIONAL EVALUATION REINFORCEMENT LEARNING
SN - 972-8865-12-0
AU - Okada H.
PY - 2004
SP - 249
EP - 255
DO - 10.5220/0001136602490255