3D UAV path optimization using a task-allocation and archive-guided mutation particle swarm optimization algorithm - Scientific Reports

Three-dimensional flight-path planning for unmanned aerial vehicles involves multiple, often conflicting objectives: minimising route length and energy consumption, maximising safety by avoiding no-fly zones and high-turbulence regions, and keeping the aircraft manoeuvrable within its kinematic limits. A study published in Scientific Reports presents an enhanced adaptation of the Task Allocation and Archive-Guided Mutation Particle Swarm Optimization algorithm, or TAMOPSO, to address them together.

Splitting the swarm into specialists

In the proposed framework, each candidate path is encoded as a sequence of discrete 3D waypoints. Rather than running one undifferentiated search, dynamic task allocation partitions the swarm into role-specific subpopulations: global explorers for broad route discovery, local refiners for optimisation close to obstacles, and altitude managers for adjusting the vertical profile.

An external archive of nondominated solutions, maintained through a uniform contribution index, preserves diversity along the Pareto front. That archive then guides adaptive Lévy-flight mutations, which shift the balance between global exploration and local refinement according to how far the swarm has converged.

What the benchmarks showed

The algorithm was evaluated across benchmark terrains of varying complexity, including scenarios with seven and ten peaks, obstacle clusters and restricted airspace constraints, and compared against fourteen baseline algorithms.

The authors report up to a 60% reduction in total flight distance and 47% faster convergence on average, alongside smoother trajectories and more uniformly distributed Pareto fronts. Extensive loop-based Monte Carlo experiments run in MATLAB were used to check the statistical robustness, computational efficiency and scalability of the approach.

Comparing behaviour across simple and complex terrain is where the difference shows. In a straightforward environment, TAMOPSO converges on the optimum along a smoother, more direct path, while competing algorithms — GA-PSO, Improved-PSO, GA, a LoRa-based method and Adapted-RRT — move more indirectly, indicating a weaker balance between exploration and exploitation.

In rugged landscapes with many local optima, the gap widens. Several of the baseline algorithms converge prematurely or oscillate, while TAMOPSO maintains steadier progress toward the global optimum. That resistance to local minima is the property that matters for real missions, where terrain and airspace restrictions rarely produce a clean search landscape.

The authors conclude that combining role-based task allocation with archive-driven adaptive mutation makes TAMOPSO a viable framework for real-time UAV path optimisation in complex 3D mission environments. The work was carried out at the University of Tabriz.