Goal-Directed Navigation with A Hybrid Planning Strategy Through Learning from Vision

Abstract

The existing hippocampal modeling approaches rarely span the wide functionality range from processing raw sensory signals to planning and action. This paper presents a goal-directed navigation system consisting of two planning strategies. The first one is a biologically inspired neural planning and navigation model that is related to learned representations of place and HD cells. It is responsible for generating spatial trajectories leading to the neighboring area of the target. The place and HD cells are trained unsupervisedly from visual images using a modified slow feature analysis (SFA) algorithm. To interpret their functional role in navigation, a planning network is trained to predict the neural activities of place and HD cell representations given selected action signals. Recursive prediction and optimization of the action signals generate goal-directed activation sequences, in which the continuous states and action spaces are represented by the population of place-, HD- and motor neuron activities. Furthermore, a second planning strategy relying on visual recognition is proposed and performs target-driven reaching on a local scale for finer accuracy. Experimental results show the effectiveness of the proposed system.