A navigational model
of spatial reference frames based on neurological and behavioral findings in human and rodents

Timo Oess
April 2016

Introduction

Spatial orientation and navigation in environments requires the agent to:

  • relate the perceptual information to previously stored memory of the same environment
  • retrieve stored spatial memory and transform it, such that it is useful in the current context
  • combine the various sensory information to infer its current position and plan a path through space

Involved Brain Regions

What are reference frames?

  • Reference frames provide a means of representing the locations of entities in space
  • They describe the orientation and distance from one entity to another and to a fixed reference point in space
  • These refernce points can change. Transformation between different frames needs to be possible.
A reference frame provides a framework in which entities can be represented

Spatial Navigation (realised in our model)

Source: www.playbuzz.com

Model Architecture

Hippocampus

Retrosplenial Cortex

Posterior Parietal Cortex

Neuron Example

Simulation Experiments

Source: www.playbuzz.com

Random Light Experiment    (Wilber et. al 2014)

- Setup

Random Light Experiment    (Wilber et. al 2014)

- Results

Random Light Experiment    (Wilber et. al 2014)

- Results

Starmaze Experiment    (Igloi et. al 2009)

- Setup

Starmaze Experiment    (Igloi et. al 2009)

- Results

Starmaze Experiment    (Igloi et. al 2009)

- Results

Starmaze Experiment    (Igloi et. al 2009)

- Results

Starmaze Experiment    (Igloi et. al 2009)

- Results

Starmaze Experiment    (Igloi et. al 2009)

- Results

Vistaspace Experiment

- Setup

  • Evaluating different strategies of the agent
  • 9 vista spaces
  • several cues located in environment

Vistaspace Experiment

- Results (unblocked environment)

Vistaspace Experiment

- Results (unblocked environment)

Vistaspace Experiment

- Results (blocked environment, uninformed agent)

Vistaspace Experiment

- Results (blocked environment, uninformed agent)

Vistaspace Experiment

- Results (blocked environment, informed agent)

Vistaspace Experiment

- Results (blocked environment, informed agent)

Conclusion

  • The simulated agent was able to exhibit similar neuronal activity as in rats in Wilber et al.
  • ➡ The system is biologically plausible.
  • We could explain switches of strategies by means of a confidence system that is continuously updated by the agent, based on the suggested movement of each strategy. Therefore we can also support the argument for simultaneously maintained strategies.
  • ➡ Supports the argument of Iglio et al. of simultaneously maintained navigation strategies.

Conclusion

  • We could show that the agent’s overall performance increased when using the route-centric strategy between vista spaces. However if a path is blocked the agent can benefit from the ability of applyingother strategies.
  • It made it flexible to react to sudden changes even without an implemented learning system
  • ➡ It seems reasonable that our agent implicitly used a hierarchical memory system of spaces

Thank you.

Allocentric Frame

Egocentric Frame

Routecentric Frame

Starmaze Experiment    (Igloi et. al 2009)

- Setup