Online conference

Sequential Planning and Spatial Navigation Through Active Inference
Giovanni Pezzulo
Team leader Cognition in Action Lab, Istituto di Scienze e Tecnologie della Cognizione, Roma

Abstract
Humans and other animals can form sequential plans, whether selecting a series of destinations during travel or assembling the steps needed to achieve a goal. In this talk, I will present studies that use the active inference framework to advance our understanding of how such sequential plans are formed and deployed in spatial navigation and other cognitive tasks that require flexible, multi-step behavior.
I will begin with simple simulations of sequential planning, illustrating how active inference accommodates both pragmatic drives such as maximizing expected utility and epistemic drives aimed at reducing uncertainty. These examples will also highlight how the underlying inferential mechanisms may correspond to neural computations. I will then turn to more complex scenarios that require integrating higher-level plans (for example, sequences of navigation goals and subgoals) with lower-level action policies (for example, step-by-step behaviors needed to reach each subgoal). To address these challenges, I will present active inference models grounded in hierarchical generative architectures that connect cognitive maps across multiple levels of abstraction, spanning both abstract “task space” and concrete “physical space,” and offering insights into interactions between frontal and hippocampal systems.
By bringing together computational modeling and neurobiological evidence, this work sheds light on how the brain constructs and executes flexible, goal-directed plans.

Invited by Neuroschool and Nicole Kolodziej, PhD Student Inmed
Monday, January 12th at 11h