Global Route Selection using Local Visual Information

When the structure of the environment is unknown, humans must navigate using local visual information. One strategy involves minimizing the angular deviation of one’s heading from a distal goal (θ). Others include minimizing the local distance (d) or turning angle (γ) to available routes. We investigated whether these variables interact to influence navigational decisions, as previously observed by Baxter & Warren (2020) for routes around a barrier. Participants walked to a goal pole through a virtual environment (32’x32’) viewed in a Quest Pro VR headset. The environment contained 3 parallel walls (“layers”), each with two doorways, yielding three binary choices per trial. Door placement was randomized to produce 64 novel configurations (half mirror-reversed), each visited once. In Experiment 1 (N=17), the goal was always visible above the walls. Experiment 2 (N=17) was identical, except that the goal disappeared before walking began. Logistic regression analyses revealed that subjects used all three local variables, minimizing deviation angle (θ), distance (d) and turn angle (γ) when selecting a doorway in each layer (all p <.01). The influence of d and θ increased with goal proximity, with θ dominating in the middle layer and d at the end. Although the goal’s disappearance weakened the θ strategy (p < .05), presumably due to spatial updating error, the other variables were constant across experiments (ns). To estimate the global consequences of local strategies, we measured the energetic cost of humans walking all possible routes and compared the performance of simulated agents following different strategies. We found that the agent that minimized θ alone selected energetically optimal routes roughly as often as the regression model, while both performed better than the d and γ agents. Our results suggest that humans navigate using a flexible local strategy that incorporates multiple variables and yields efficient global routes.