Composing trajectories for rapid inference of navigational goals

Area of Science:

• Computational neuroscience
• Animal behavior
• Robotics

Background:

• Animals exhibit remarkable spatial navigation efficiency, localizing targets in few trials.
• Existing models struggle to explain this rapid learning, requiring extensive experience.

Purpose of the Study:

• To develop a computational model explaining efficient spatial learning in mice.
• To understand how animals generate structured trajectories for target interception.

Main Methods:

• Designed agents controlling speed and angular velocity between anchor points.
• Utilized Bayesian inference and active sampling for rapid anchor learning.
• Modeled agent behavior on observed mouse navigation strategies.

Main Results:

• Agents learned to intercept hidden targets within tens of trials.
• The model captured behavioral structure evolution and learning efficiency limits.
• Algorithm explained obstacle avoidance and adaptation to target switches.

Conclusions:

• The proposed algorithm provides a framework for understanding efficient animal navigation.
• It integrates egocentric and allocentric strategies naturally.
• This model advances our understanding of learning and decision-making in spatial environments.