Adaptive Artificial Potential Field Integrated with World Model Logic for Autonomous UAV Reconnaissance in Smoke Conditions

Autonomous UAV reconnaissance in smoke-obscured environments faces dual challenges: increased collision risk and degraded target detection due to reduced optical sensor visibility. This study proposes an adaptive Artificial Potential Field(APF) method integrating a probabilistic World Model(WM) logic with a Cumulative Detection Probability(CDP) model. The WM logic, derived from Thrun et al.'s occupancy grid framework, retains estimated obstacle positions during sensor dropout events, preventing collisions. The CDP model, grounded in Koopman's continuous search theory, redefines target detection as a time-integrated stochastic process. A visibility-adaptive controller actively adjusts repulsive gain and flight speed in response to real-time visibility. Monte Carlo simulations (N = 200) and 2×2 factorial analysis demonstrate that the proposed WM+APF achieves a survival rate of 82.5 % and detection rate of 81.4 %(H-mean: 81.9), significantly outperforming conventional APF(36.0 % / 63.4 %). Factorial analysis confirms WM logic as the dominant factor(+47.2 %p survival), with visibility adaptation providing consistent incremental gains.