Adaptive Strategies and Anticipatory Timing in Pedestrian Collision Avoidance: A Systematic Review

다국어 초록 (Multilingual Abstract)

Background : Pedestrian collision avoidance is an anticipatory, perception–action process shaped by spatial geometry, social context, and sensory–attentional constraints. Yet evidence on how (strategy) and when (timing) avoidance is initiated has been dispersed across paradigms.
Methods : Following PRISMA 2020, we synthesized 14 empirical studies (2005–2025) spanning controlled lab, virtual reality, and ecological settings that reported quantitative or qualitative outcomes on avoidance strategy and/or initiation timing during human walking.
Results : Two mechanisms consistently dominated: directional adjustment (steering) and speed modulation. Steering was the default response across dyadic and oblique encounters, while speed modulation occurred preferentially for yielding roles, acute angles, or constrained space, and combined responses arose when residual risk persisted. Initiation was anticipatory, commonly occurring when the Minimum Predicted Distance (MPD) dropped below a small threshold (≈<1 m), with MPD trajectories showing observation, reaction, and regulation phases, indicating proactive control rather than reflexive responses. In multi-agent contexts, pedestrians typically resolved conflicts sequentially, although certain formations invited simultaneous co-adaptation, as revealed by joint analysis of MPD and dynamic gap affordances. Under path constraints, walkers recruited body rotation, which scaled with corridor width and became critical for safe passing in narrow spaces. Ecological data showed robust role asymmetries: singles bore greater avoidance effort, and higher dyad interaction weakened dyads’ responsiveness; deviation magnitude scaled with social-interaction level even at comparable risk.
Conclusions : Evidence converged on a context–strategy–timing framework: low MPD triggers avoidance; context (geometry, role, social engagement, constraint) selects the mix of steering and speed control; and postural freedom (rotation) engage as space tightens. This synthesis offers mechanistic targets for predictive pedestrian models, human–robot navigation policies, and rehabilitation programs that train timely steering first, with speed and rotation layered under elevated risk or constraint.