Intelligent systems and robots

A biologically plausible model for the construction of a cognitive agent suboptimal path

E. E. Parfenova, I. H. Shungarov, G. A. Kiselev

Peoples' Friendship University of Russia named after Patrice Lumumba, Moscow

Abstract: The article describes a biologically plausible hybrid navigation algorithm for an autonomous mobile agent in an unidentified dynamic environment. The algorithm is based on the discrete optimization method and extended by an incremental descriptor. The incremental descriptor is a statistical memory module that implements the step-by-step (incremental) accumulation and updating of information about the relationships between states and actions. It acts as a probabilistic approximation of the environment transition function, which improves the robustness of training and enables a balance between exploration and exploitation without explicitly modeling the dynamics. It is a probabilistic memory and accumulates statistics of state–action pairs. Consequently, the learning stability is improved by balancing between “exploration” and “exploitation” without explicitly modeling transitions in the environment.

Keywords: reinforcement learning, discrete optimization, bayesian descriptor, potential fields, robot navigation, cognitive agents.

DOI: 10.14357/20718594250404

Bibliographic databases:

•