A neural field approach to robot navigation with brain-inspired goal-directed cognitive maps

Tools

Hicks, Matthew and Lei, Tingjun and Sellers, Timothy and Luo, Chaomin and Yang, Erfu; (2026) A neural field approach to robot navigation with brain-inspired goal-directed cognitive maps. In: 2025 IEEE International Conference on Robotics and Biomimetics (ROBIO). 2025 IEEE International Conference on Robotics and Biomimetics (ROBIO) . IEEE, CHN, pp. 269-276. ISBN 979-8-3315-5747-8 (https://doi.org/10.1109/robio66223.2025.11377267)

[thumbnail of Hicks-etal-2026-A-neural-field-approach-to-robot-navigation-with-brain-inspired-goal-directed-cognitive-maps]
Preview
 Text. Filename: Hicks-etal-2026-A-neural-field-approach-to-robot-navigation-with-brain-inspired-goal-directed-cognitive-maps.pdf
Accepted Author Manuscript
License: Creative Commons Attribution 4.0 logo
Download (794kB)| Preview

Abstract

Mammals possess remarkable navigational abilities, underpinned by the hippocampus and associated neural cells that enable them to orient, adapt, and reach goals in unfamiliar environments without exhaustive exploration. Traditional robotic navigation systems, in contrast, depend often on static maps or pre-programmed routes, making them vulnerable to dynamic changes, incomplete information, and local minima. To bridge this gap, this paper proposes an enhanced goal-directed cognitive map (eGDCM) model that integrates a neural field-based algorithm for adaptive optimization of head direction and speed cells, augmented by the follow-the-gap method (FGM) for safe motion, and improved Vector Field Histogram (iVFH) mapping. The eGDCM fuses the spatial representation and goal-driven reasoning capabilities of cognitive maps with the continuous, self-organizing dynamics of neural fields, enabling real-time adaptation to both static and dynamic obstacles. This braininspired framework incorporates multiple navigational strategies allowing robots to form compact, efficient spatial graphs and plan collision-free paths without computationally intensive graph searches. Simulation and comparison studies on complex city maps and dynamic environments demonstrate that eGDCM achieves shorter and smoother trajectories, and superior adaptability compared to the GDCM model and other state-of-the art algorithms. These results highlight the potential of hippocampalinspired neural mechanisms to advance robust, efficient, and safe autonomous navigation in unstructured real-world settings.

ORCID iDs

Hicks, Matthew, Lei, Tingjun, Sellers, Timothy, Luo, Chaomin and Yang, Erfu ORCID logoORCID: https://orcid.org/0000-0003-1813-5950;
CORE (COnnecting REpositories)