Cue Integration in Spatial Navigation – a Bayesian Approach

Abstract

The current study aimed to apply the Bayesian cue integration paradigm to examine spatial cue interaction in human navigation. Participants completed a triangle completion task either in an immersive virtual environment using self-motion cues and visual landmarks (Experiments 1 to 3) or on a 2D desktop virtual display using proximal and distal landmarks (Experiment 4). Experiments 1A and 1B extended previous studies by experimentally manipulating cue reliability of self-motion cues and visual cues independently and within-subjects. Results showed that participants integrated these two cue types optimally in a Bayesian manner overall, but at the cost of being sub-optimal at individual reliability levels of the manipulated cue. Experiment 2 was designed to mislead participants about cue reliability by providing distorted feedback. Cue relative reliability was changed by the feedback, and the direction of change depended on how the feedback was distorted. Despite changes in cue reliability, Bayesian principles mostly held true, but an additional hysteresis assumption was needed to fully explain the results. Experiment 3 aimed to incorporate landmark instability into the paradigm. The results were directionally consistent with the hypothesis that cue instability affected cue weights indirectly by influencing cue reliability. In all three of these experiments, we observed positive correlations across participants between actual weights assigned to the cues and Bayesian weights, demonstrating that the cue weighting process in general followed Bayesian principles when self-motion cues and visual cues were concerned. Experiment 4 represents our efforts to accommodate the paradigm to a 2-dimensional virtual display on desktop computer in order to examine the interaction between different visual cues. We examined proximal and distal landmarks, whose interaction largely deviated from Bayesian principles. Together, the findings from our 5 experiments help us to gain a deeper understanding of how spatial cues interact during the process of spatial navigation. Our findings not only help clarify some critical issues in the spatial navigation domain but also impose more questions for future investigation.