Neural Networks and Q-Learning for Robotics

Claude Touzet (1)

1 LNIA - Laboratoire de Neurosciences intégratives et adaptatives

Introduction Behavior-Based Approach Supervised Learning of a Behavior Miniature Mobile Robot Khepera Illustration: Reward-Penalty Learning Reinforcement Learning Genetic Algorithms Learning Classifier Systems GA & ANN Q-learning Evaluation Function Algorithm Reinforcement Function Update Function Convergence Limitations Generalization Neural Implementations of the Q-learning Multilayer Perceptron Implementation (ideal & Q-CON) Q-KOHON Comparison Knowledge Incorporation Reinforcement Function Design Building of a non-explicit Model Learning in Cooperative Robotics References

Mots clés: Reinforcement learning; robotics

Domaines: Intelligence artificielle [cs.AI]

Claude Touzet. Neural Networks and Q-Learning for Robotics. IJCNN’99 (International Joint Conference (IEEE INNS) on Neural Networks), Jul 1999, Washington DC, United States. ⟨hal-01355056⟩