Embodied Cognition-Inspired Robot Navigation Using Neural Field Dynamics And Perceptual Binding

Abstract

This paper presents a neural-field-based framework for embodied cognition-inspired navigation in mobile robots, integrating dynamic neural field dynamics with a perceptual binding mechanism using 2D LiDAR sensing. Unlike conventional navigation pipelines that rely on discrete mapping or symbolic representations, the proposed approach maintains continuous field-based representations across perception, spatial memory, and motor control, enabling direct sensorimotor coupling in real time. The architecture combines continuous attractor neural networks with biologically-inspired binding operations to support adaptive behaviour and robust navigation under changing environmental conditions.

A modular ROS1 implementation is developed and validated on TurtleBot3 platforms through extensive real-robot experiments conducted across static, dynamic, and cluttered environments. Experimental results demonstrate reliable performance, achieving a 94.7% overall navigation success rate while maintaining real-time execution with average processing latency below 50 ms at a 20 Hz update rate. The perceptual binding mechanism improves integration of sensory input and spatial memory, producing adaptive behaviours such as context-dependent speed modulation, predictive path selection, and obstacle avoidance without explicit symbolic planning.

The results show that neural field dynamics provide an effective computational framework for biologically-inspired sensorimotor control, combining behavioural adaptability with computational efficiency suitable for embedded robotic deployment.