Main Menu (Mobile)- Block
Main Menu - Block
Abstract
Natural organisms navigate a world in constant flux and are frequently confronted with sensory ambiguity and shifting goals. While biological systems can adapt rapidly to navigate in novel and nonstationary environments, this challenge remains a significant hurdle for artificial intelligence. In this review, we examine recent progress in understanding neurobiological systems that allow for adaptive, efficient spatial and abstract navigation in dynamic environments and how these observations have informed new developments for the same behavior in machines. We highlight how predictive learning serves as a unifying framework to describe the organization of neural activity across distributed systems for navigation, including the hippocampal formation, parietal cortex, and higher-order associative structures. Specifically, we explore how such neurobiological systems encode expected future states, dynamic target trajectories, action plans, and internal behavioral strategies. Finally, we contrast the diffuse, inefficient exploration of canonical artificial agents with the structured, efficient control policies for goal-directed navigation observed in natural organisms. By grounding learning in the interplay between action, prediction, and natural behavioral statistics, navigation provides a roadmap for building resilient, neurobiologically inspired continual learning in artificial agents.
