Almost there: Learning to navigate approximately with a grid map

Area of Science:

• Animal behavior
• Computational neuroscience
• Spatial cognition

Background:

• Grid map navigation relies on environmental gradients for spatial orientation.
• Mechanisms for extrapolating and combining gradient information remain unclear.
• The developmental trajectory of these navigational strategies is poorly understood.

Purpose of the Study:

• To investigate the ontogeny of grid map navigation mechanisms.
• To determine if navigational strategies are innate or learned.
• To model gradient extrapolation and combination in artificial neural networks.

Main Methods:

• Trained artificial neural networks in virtual bicoordinate grid environments.
• Tested network outputs against existing models of spatial gradient use.
• Analyzed how networks adapted their navigation strategies during training.

Main Results:

• Networks initially adopted an 'approximate model' for navigation.
• This model involves independent gradient assessment and summation of displacements.
• Networks evolved beyond the approximate model through learning, adapting gradient extrapolation based on the environment.

Conclusions:

• The approximate model may represent an early-stage navigational strategy.
• Navigational mechanisms are not fixed and change through learning.
• Environmental context influences how animals learn to use spatial gradients for navigation.