AI-driven path optimization of virtual-target-based reactive collision avoidance algorithm using genetic algorithms

In this paper, the previously proposed virtual-target (VT) approach for reactive collision avoidance and pathfinding in autonomous unmanned aerial vehicles (UAVs) is integrated with an artificial intelligence-driven optimization framework. While the VT-based method enables local navigation by locating suitable open gaps in the UAV’s sensor field to avoid obstacles, its effectiveness is significantly influenced by two control parameters: 1. The distance at which avoidance movements must be started, the threshold distance dthr and 2. The weighting factor u, which controls how aggressively direction adjustments are made by the UAV. To minimize the likelihood of collisions and UAV trajectory path length during low-altitude flight in unstructured environments, this work optimizes these parameters using a genetic algorithm (GA). The GA systematically adjusts parameter values through fitness evaluation, selection, crossover, and mutation. It also uses method of elitism to keep the best-performing solutions of the previous generations. Simulation findings demonstrate that the GA-optimized VT-based method converges at optimal parameter vector of (dth=3m,u=0.651), yielding a candidate trajectory with a maximum fitness value of 5.4 × 10−5. Furthermore, the convergence of the average fitness towards a closely matching value of 5.2 × 10−5 confirms healthy convergence and population wide improvement over the generations. This eventually resulted in a smoother trajectory, collision less path and efficient navigation when compared to non-optimized configuration. The findings show that optimization significantly increases the robustness and reliability of the reactive obstacle avoidance, providing a promising step toward the practical deployment of UAVs in unstructured and unknown environments.