Autonomous agent based on reinforcement learning and adaptive shadowed network

摘要: Abstract The planning of intelligent robot behavior plays an important role in the development flexible automated systems. robot’s intelligence comprises its capability to act unpredictable and chaotic situations, which requires not just a change but creation working knowledge. Planning addresses three main issues: finding task solutions unknown learning from experience recognizing similarity problem paradigms. This article outlines system integrates reinforcement method neural network approach with aim ensure autonomous conditions. assumption is that tabula rasa has no knowledge work space structure. Initially, it basic strategic searching for solutions, based on random attempts, built-in system. used here evaluate induce new, or improve existing, acquired action (task) plan stored as can be solving similar future problems. To provide recognition similarities, Adaptive Fuzzy Shadowed designed. novel concept fuzzy rule shadowed hidden layer architecture enables slightly translated rotated patterns does forget already learned structures. simulated using object-oriented techniques verified planned examples, proving advantages proposed approach: learning, invariant regard order training samples, single iteration progress.