Improved Harris Hawks Optimization Refines Three-Dimensional Drone Flight Path Planning

In an effort to overcome the computational and safety bottlenecks that hinder automated flight navigation across complex terrains, an engineering research team has introduced an enhanced trajectory planning framework based on an Improved Harris Hawks Optimization algorithm. Autonomous unmanned aerial vehicles operating in three-dimensional environments must routinely navigate between spatial coordinates while balancing battery endurance, aerodynamic stability, and obstacle proximity. By formulating a multi-constraint weighted-cost objective function that holistically penalizes inefficient transit while incorporating a dynamic time-to-collision safety metric, the researchers demonstrated in numerical simulations that their modified meta-heuristic consistently finds smoother, shorter, and collision-free flight corridors where conventional algorithms frequently stall or fail.

The fundamental challenge of three-dimensional path planning for unmanned aerial vehicles lies in the mathematically complex, non-deterministic polynomial-time hard character of continuous spatial optimization. Battery-powered aircraft possess finite energy reserves, which makes direct flight distance a primary determinant of mission viability. However, simply choosing the shortest spatial line between a designated starting point and an arrival destination is rarely feasible in realistic operational theaters. Low-altitude flight paths must negotiate irregular elevation profiles, avoid natural landforms, and maneuver around artificial structures or restricted airspace corridors. While traditional global routing strategies such as Dijkstra’s algorithm, the A* search method, rapidly exploring random trees, and artificial potential fields have seen widespread application, they routinely encounter severe limitations in dense three-dimensional spaces, including prohibitive computer memory overhead, slow convergence rates, and chronic vulnerability to becoming trapped in locally optimal but globally suboptimal solutions.

To bypass the rigid scaling limitations of classical deterministic and grid-based algorithms, contemporary robotics researchers have increasingly turned to swarm intelligence techniques, which emulate the decentralized cooperative mechanisms observed in biological populations. Formulations utilizing particle swarm optimization, genetic algorithms, ant colony systems, grey wolf optimizers, and whale optimization methods have achieved varying degrees of success in generating viable aerial paths. Nevertheless, as the dimensionality and kinematic constraints of the flight envelope expand, standard bio-inspired algorithms exhibit systematic performance degradation. They frequently struggle to navigate multi-modal optimization landscapes, exhibiting sluggish convergence profiles and failing to effectively balance competing physical metrics such as minimizing path length while dampening violent attitude adjustments. The conventional Harris Hawks Optimization framework, originally inspired by the cooperative pouncing and pursuit tactics of raptor groups, suffers from identical structural drawbacks, notably limited population diversity during early search iterations and an inflexible linear progression from global exploration to localized exploitation.

To address these shortcomings, the researchers designed an integrated cost function that compiles five distinct mission metrics into a single weighted mathematical objective. The first component assesses overall trajectory length, calculated across a discrete sequence of waypoints, which directly serves as an analytical proxy for flight duration, actuator energy expenditure, and battery consumption under steady cruising speeds. The second component governs flight altitude, introducing a root-mean-square penalty formulation that confines the aircraft within a designated vertical safety corridor above underlying topography while suppressing disruptive vertical oscillations. The third metric imposes kinematic attitude boundaries on horizontal yaw deflection and vertical pitch variation, penalizing excessive angular deviations to guarantee that the planned path remains within the structural maneuverability limits of the airframe. The final two components incorporate static spatial threat distances and a dynamic obstacle avoidance response time based on the time-to-collision principle, which evaluates instantaneous approach velocities to ensure adequate time margins for emergency avoidance maneuvers.

The operational engine of this path planning architecture is the Improved Harris Hawks Optimization algorithm, which integrates three targeted structural modifications to remedy the limitations of the baseline heuristic. The first enhancement introduces a nature-inspired position update and trap-avoidance mechanism that calculates repulsive forces away from the worst-performing candidate solutions while injecting stochastic directional variations to augment swarm diversity during the exploratory phase. The second modification replaces the traditional linear energy dissipation model with a cosine-exponential nonlinear energy decay schedule governed by a dedicated decay parameter. This nonlinear profile enforces a gradual initial decline to preserve broad exploratory search, transitions into a steep descent during intermediate iterations to accelerate convergence, and asymptotically levels off at a non-zero terminal threshold to retain sufficient residual mobility for escaping late-stage local minima. The third improvement modulates the local exploitation phase by executing ten consecutive Lévy flight iterations during dive maneuvers, refining solution precision near promising corridors.

The computational workflow was validated through comprehensive simulation experiments implemented on the MATLAB R2021b platform running on a dedicated workstation equipped with an Intel Core i9 processor and 64 gigabytes of memory. The experimental scenario modeled a complex three-dimensional terrain spanning a horizontal operational domain where an aircraft navigates from an initial waypoint coordinate at (10, 90) to a target destination at (130, 10). The planned trajectory was discretized into 150 intermediate waypoints with altitudes determined via terrain interpolation. To simulate realistic environmental threats and restricted airspace, eight cylindrical obstacles of varying radii ranging from 5 to 10 meters were deployed across the map, with protective horizontal influence buffers configured at 0.2 meters beyond each obstacle’s physical perimeter. Optimization parameters were established with a swarm population size of 50 individuals, a maximum limit of 200 iterations, and an objective weighting distribution assigning 0.5 to trajectory length, 0.3 to altitude control, and 0.2 to flight attitude smoothness.

Across 30 independent experimental trials conducted on the test map, the proposed method converged reliably to high-quality flight profiles, achieving a mean total weighted cost of 81.3020 with a standard deviation of 0.8548, while recorded static and dynamic safety penalty costs remained strictly at zero across all runs. In an isolated optimal single-run execution, the algorithm achieved a total cost value of 79.5070, corresponding to an absolute path length cost of 154.6100, an altitude deviation cost of 1.0297, and an attitude cost of 9.4656. In direct statistical comparisons against the baseline standard Harris Hawks Optimization algorithm, which recorded an average total cost of 86.0575, the improved variant demonstrated marked superiority across all evaluated metrics. Specifically, the enhanced method reduced average path length cost from 167.4558 down to 158.1035, diminished vertical altitude variation from 1.2698 to 1.0607, and lowered attitude cost from 9.7435 to 9.6606, confirming that the algorithmic enhancements consistently produce shorter paths that adhere more tightly to terrain contours without compromising flight smoothness.

To rigorously verify the individual and combined efficacy of each algorithmic enhancement, the authors conducted comprehensive ablation trials comparing eight distinct module configurations across 30 runs each. Interestingly, variants incorporating the modulated Lévy flight iterations without full diversity safeguards exhibited bimodal convergence behaviors, wherein certain runs achieved competitive costs near 80.4 while others became trapped at suboptimal plateaus around 85.3. Only the fully integrated framework, combining the diversified position updates, nonlinear energy scheduling, and ten Lévy iterations, achieved stable convergence to the superior cost tier in all 30 trials, proving that early-stage exploratory dispersion and late-stage localized refinement operate in necessary structural synergy.

Benchmarking against six prominent contemporary meta-heuristics further substantiated the competitive standing of the proposed approach. While the Dung Beetle Optimizer achieved the lowest attitude penalty of 9.1424, its higher path length and altitude costs resulted in an overall fitness of 83.1395, which remained 2.2 percent higher than the proposed method’s 81.3020. Other algorithms such as the Newton-Raphson-Based Optimizer and Non-Continuous Harris Hawks Optimization exhibited severe convergence limitations, producing path length costs exceeding 180 and total costs reaching 95.6595 and 102.7601. In terms of computational overhead, the proposed method recorded an average execution runtime of 168.28 seconds compared to 121.60 seconds for the baseline algorithm, representing a 38.4 percent processing increase that remains within acceptable bounds given the asymptotic complexity preservation of order N by T by D.

Parametric sensitivity analyses confirmed that the algorithm’s performance gains remain robust across varying operational configurations. Testing different Lévy flight iteration numbers indicated that setting the refinement counter to ten was essential to eliminate bimodal branching without incurring the superfluous computational costs associated with higher values. Similarly, evaluating the energy decay parameter demonstrated that moderate values between 0.5 and 2.0 sustained balanced exploration-exploitation transitions, whereas excessive values above 2.5 prematurely suppressed terminal residual energy and increased solution dispersion. Systematic adjustments across nine distinct objective weighting distributions further verified that path length and attitude metrics remained tightly bounded within narrow margins regardless of prioritization shifts. Although these findings substantiate the simulated feasibility of the multi-constraint optimization model, the researchers noted key operational boundaries, emphasizing that future development must address real-time trajectory re-planning in the presence of dynamic moving hazards and validate algorithmic execution on physical drone avionics hardware.

Subject of Research: Three-dimensional unmanned aerial vehicle path planning using an improved Harris Hawks optimization algorithm

Article Title: Multi-constraint weighted-cost optimization for UAV path planning using an improved harris hawks optimization

Article References: Peng, Y., Zhao, Y., He, W., Jiang, B., Wu, B., & Zhou, D. (2026). Multi-constraint weighted-cost optimization for UAV path planning using an improved harris hawks optimization. Frontiers in Robotics and AI, 13, 1979434. https://doi.org/10.3389/frobt.2026.1979434

Image Credits: AI Generated

DOI: 10.3389/frobt.2026.1979434

Keywords: Unmanned Aerial Vehicles, Path Planning, Harris Hawks Optimization, Swarm Intelligence, Trajectory Optimization, Obstacle Avoidance, Flight Altitude Control, Flight Attitude Constraints, Kinematic Constraints, Time-to-Collision, Lévy Flight, Evolutionary Computation, Three-Dimensional Mapping, Meta-Heuristic Algorithms, Flight Efficiency, Robotics Simulation, Dynamic Risk Assessment, Ablation Analysis