A self-organizing neural architecture integrating desire, intention and reinforcement learning

作者：Tan, Ah-Hwee^*; Feng, Yu-Hong; Ong, Yew-Soon

来源：Neurocomputing, 2010, 73(7-9): 1465-1477.

DOI：10.1016/j.neucom.2009.11.012

• Reinforcement learning
• Plan learning
• Self-organizing neural networks
• BDI
• Minefield navigation

This paper presents a self-organizing neural architecture that integrates the features of belief, desire, and intention (BDI) systems with reinforcement learning. Based on fusion Adaptive Resonance Theory (fusion ART), the proposed architecture provides a unified treatment for both intentional and reactive cognitive functionalities. Operating with a sense-act-learn paradigm, the low level reactive module is a fusion ART network that learns action and value policies across the sensory, motor, and feedback channels. During performance, the actions executed by the reactive module are tracked by a high level intention module (also a fusion ART network) that learns to associate sequences of actions with context and goals. The intention module equips the architecture with deliberative planning capabilities, enabling it to purposefully maintain an agenda of actions to perform and to reduce the need of constantly sensing the environment. Through reinforcement learning, plans can also be evaluated and refined without the rigidity of user-defined plans. We examine two strategies for combining the intention and reactive modules for decision making in a real time environment. Our experiments based on a minefield navigation domain show that the integrated architecture is able to learn plans efficiently, achieve good plan utilization, and combine both intentional and reactive action execution to yield a robust performance.

• 出版日期2010-3
• 单位中国科学院深圳先进技术研究院