SIMULTANEOUS LEARNING OF PERCEPTIONS AND ACTIONS IN AUTONOMOUS ROBOTS

Pablo Quintía, Roberto Iglesias, Miguel Rodríguez, Carlos V. Regueiro

Abstract

This paper presents a new learning approach for autonomous robots. Our system will learn simultaneously the perception – the set of states relevant to the task – and the action to execute on each state for the task-robotenvironment triad. The objective is to solve two problems that are found when learning new tasks with robots: interpretability of the learning process and number of parameters; and the complex design of the state space. The former was solved using a new reinforcement learning algorithm that tries to maximize the time before failure in order to obtain a control policy suitable to the desired behavior. The state representation will be created dynamically, starting with an empty state space and adding new states as the robot finds them, this makes unnecessary the creation of a predefined state representation, which is a tedious task.

References

  1. Carpenter, G. A., Grossberg, S., and Rosen, D. B. (1991). Fuzzy art: Fast stable learning and categorization of analog patterns by an adaptive resonance system. Neural Networks, 4(6):759-771.
  2. Iglesias, R., Regueiro, C. V., Correa, J., Sánchez, E., and Barro, S. (1998). Improving wall following behaviour in a mobile robot using reinforcement learning. In ICSC International symposium on engineering of intelligent systems (EIS'98), Tenerife (Espan˜a).
  3. Nehmzow, U., Iglesias, R., Kyriacou, T., and Billings, S. (2006). Robot learning through task identification. Robotics and Autonomous Systems, 54:766-778.
  4. Sutton, R. S. and Barto, A. G. (1998). Reinforcement Learning: an Introduction. MIT Press.
  5. Watkins, C. (1989). Learning from delayed rewards. PhD thesis, University of Cambridge, England.
Download


Paper Citation


in Harvard Style

Quintía P., Iglesias R., Rodríguez M. and V. Regueiro C. (2010). SIMULTANEOUS LEARNING OF PERCEPTIONS AND ACTIONS IN AUTONOMOUS ROBOTS . In Proceedings of the 7th International Conference on Informatics in Control, Automation and Robotics - Volume 2: ICINCO, ISBN 978-989-8425-01-0, pages 395-398. DOI: 10.5220/0002943203950398


in Bibtex Style

@conference{icinco10,
author={Pablo Quintía and Roberto Iglesias and Miguel Rodríguez and Carlos V. Regueiro},
title={SIMULTANEOUS LEARNING OF PERCEPTIONS AND ACTIONS IN AUTONOMOUS ROBOTS},
booktitle={Proceedings of the 7th International Conference on Informatics in Control, Automation and Robotics - Volume 2: ICINCO,},
year={2010},
pages={395-398},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0002943203950398},
isbn={978-989-8425-01-0},
}


in EndNote Style

TY - CONF
JO - Proceedings of the 7th International Conference on Informatics in Control, Automation and Robotics - Volume 2: ICINCO,
TI - SIMULTANEOUS LEARNING OF PERCEPTIONS AND ACTIONS IN AUTONOMOUS ROBOTS
SN - 978-989-8425-01-0
AU - Quintía P.
AU - Iglesias R.
AU - Rodríguez M.
AU - V. Regueiro C.
PY - 2010
SP - 395
EP - 398
DO - 10.5220/0002943203950398