350 rub
Journal Neurocomputers №5 for 2013 г.
Article in number:
Training neural network for an obstacle-avoiding autonomous mobile robot using reinforcement learning
Keywords: reinforcement learning neural network control mobile robots
Authors:
R.A. Munasypov, G.A. Saitova, S.S. Moskvichev, T.R. Shakhmametev
Abstract:
In this article we present an approach to the obstacle-avoidance task in unknown environment for autonomous mobile robots using reinforcement learning neural network. Q-learning is one of reinforcement learning methods widely used in autonomous mobile robotics. This method is used to train a neural network controller of a mobile robot providing it with autonomous obstacle-avoiding behavior in unknown environment. Simulation results show that the method allows the robot to achieve an efficient locomotion strategy with no collisions with the environment.
Pages: 14-18
References

  1. Borenstein J., Koren Y. Real-time Obstacle Avoidance for Fast Mobile Robots // IEEE Transactions on Systems, Man, and Cybernetics. 1989. V. 19. № 5. P. 1179-1187.
  2. Stankevich L.A. Intellektualnye roboty i sistemy upravlenija // Nejjrokompjutery: razrabotka, primenenie. 2005. № 8-9. S. 54-66.
  3. Sutton R., Barto A. Reinforcement learning: An introduction. Adaptive Computation and Machine Learning series. Cambridge: MIT Press (Bradford Book). 1998. V. 18. 322 p.
  4. Watkins J., Dayan P. Q-learning // Machine Learning. 1992. V. 8. P. 279-292.
  5. Onat A. Q-learning with recurrent neural networks as a controller for the inverted pendulum problem // The Fifth International Conference on Neural Information Processing. October 21-23. 1998. P. 837-840.
  6. Cervera E., del Pobil A.P. Sensor-based learning for practical planning of fine motions in robotics // Information Sciences. 2002. V. 145. Iss. 1-2. P. 147-168.