Exploring unknown environments: motivated developmental learning for autonomous navigation of mobile robots

Abstract

How to realize flexible behavior decision making is an important prerequisite for mobile robots to perform various tasks. To solve the problems of poor real-time performance and adaptability of traditional methods, this paper proposes a method that simulates cerebellar function through developmental network, and simulates the function of “what” and “where” channels in the visual system as well as the neuromodulatory mechanisms of dopamine and serotonin, so as to improve the adaptability of cerebellar model to behavioral decision making under supervised learning strategies. At the same time, this paper pays special attention to the strategy of simulating cerebellar reinforcement learning. By simulating the sleep recall mechanism of hippocampus and the neuromodulatory mechanism of acetylcholine and norepinephrine, mobile robots can have continuous and stable learning ability in unfamiliar environment, and improve the real-time and adaptability of their behavioral decision making. Simulation results in both static and dynamic environments, as well as the results in the static physical environment, validate the potential of this model, indicating that the cerebellar model based on reinforcement learning plays an important role in the behavioral decision making of mobile robots.