Synthetic Spatial Foraging With Active Inference in a Geocaching Task.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Humans excel at spatial learning and navigation through flexible environmental representations.
• Goal-directed behavior and spatial foraging rely on sophisticated cognitive processes.
• Active Inference provides a framework for understanding how agents optimize beliefs about their environment.

Purpose of the Study:

• To develop a deep Active Inference model simulating human spatial learning and goal-directed behavior.
• To investigate the role of belief updating in spatial foraging and navigation.
• To explore the behavioral and neurophysiological correlates of active inference in a simulated geocaching task.

Main Methods:

• A deep Active Inference model was formulated to optimize Bayesian beliefs for maximizing model evidence.
• Simulations of a geocaching task were used to model agents learning environmental likelihoods and navigating to target locations.
• The model's belief updating mechanisms were analyzed in relation to spatial foraging and discovery of hidden objects.

Main Results:

• The Active Inference model successfully simulated goal-directed spatial foraging and navigation in a geocaching task.
• Belief updating was shown to be a key mechanism underlying the agent's ability to learn and adapt to the environment.
• The simulations provided insights into the computational principles driving spatial exploration and target localization.

Conclusions:

• Deep Active Inference offers a powerful computational framework for understanding complex behaviors like spatial navigation and foraging.
• Belief updating is crucial for agents to form and utilize flexible spatial representations of their environment.
• This model provides a foundation for further research into the neurobiology of spatial cognition and decision-making.