Bayesian models of human navigation behaviour in an augmented reality audiomaze

Area of Science:

• Cognitive Neuroscience
• Computational Modeling
• Human Navigation Behavior

Background:

• Understanding human spatial navigation is crucial for fields like robotics and cognitive science.
• Previous research suggests internal spatial representations, or maps, aid navigation.
• The precise mechanisms and benefits of map learning in real-space navigation remain an active area of investigation.

Purpose of the Study:

• To investigate the impact of different map learning strategies on human whole-body motion capture data during a real-space navigation task.
• To compare Bayesian models representing no map learning, single-trial map learning, and cumulative map learning.
• To explore how individual navigation strategies (egocentric vs. allocentric) interact with map learning.

Main Methods:

• Utilized Bayesian modeling to analyze human motion capture data from an exploratory navigation task in an Audiomaze environment.
• Developed and compared three models: feedback-only (no map learning), map resetting (limited map learning), and map updating (cumulative map learning).
• Estimated behavioral variables including step sizes and turning angles for each model and participant group.

Main Results:

• Map learning models (map resetting and map updating) consistently yielded more accurate estimates of step sizes compared to the feedback-only model.
• Turning angle estimates were also more accurate with map learning, particularly by the third trial.
• Allocentric navigators showed a significant advantage in step size estimation when using the map updating model, suggesting greater benefit from cumulative map learning.

Conclusions:

• The study provides Bayesian evidence that human map learning significantly enhances navigation behavior, improving accuracy in estimating movement parameters.
• Allocentric navigation strategies appear to leverage cumulative map learning more effectively than egocentric strategies.
• Findings offer insights into human spatial cognition and have potential implications for the simultaneous localization and mapping (SLAM) problem in artificial systems.