Predictive maps in rats and humans for spatial navigation

Area of Science:

• Comparative Cognition
• Computational Neuroscience
• Animal Behavior

Background:

• Traditional navigation studies focus on single species with task-specific designs.
• Understanding spatial navigation requires a framework that integrates diverse agents.
• Dynamic environments necessitate adaptive navigation strategies.

Purpose of the Study:

• To develop a unified experimental and analytic framework for studying spatial navigation.
• To compare navigation dynamics across humans, rats, and reinforcement learning (RL) agents.
• To identify shared mechanisms underlying adaptive navigation in changing environments.

Main Methods:

• Development of a novel open-field navigation task ('Tartarus maze') with dynamic obstructions.
• Integration of behavioral data from humans and rats with simulated RL agents.
• Analysis of trajectory data to compare navigation strategies, including successor representation and model-based RL.

Main Results:

• Human and rat navigation trajectories showed remarkable similarity in the Tartarus maze.
• Both species' behavior closely resembled RL agents employing a successor representation (predictive mapping).
• Human trajectories also aligned with model-based RL agents using tree-search planning.

Conclusions:

• Comparative analysis across species and computational models refines understanding of mammalian navigation.
• The successor representation is a key mechanism for navigation in dynamic environments.
• Integrating diverse agents highlights conserved neural and behavioral mechanisms for spatial navigation.