Hybrid Metaheuristic Algorithm Enhances Route Planning and Smoothness for Mobile Robots

Researchers have developed a dual-layer hybrid route optimization model that significantly improves how mobile robots navigate static grid environments by combining an improved Sine Cosine Algorithm with an improved Beetle Antennae Search. Published in Discover Artificial Intelligence, the study demonstrates that sequentially pairing these two metaheuristic algorithms achieves an effective balance between global exploration and local refinement. In benchmark evaluations, the integrated algorithm attained superior convergence accuracy and stability compared with existing population-based optimizers. When deployed within simulated grid-based workspaces, the hybrid model produced paths that were shorter, contained fewer sharp turns, and exhibited continuous geometric smoothness. Specifically, in a simple obstacle map, the planned route reached a total length of 29.22 metres, representing a reduction between 16.73 percent and 23.83 percent compared to competing methods, while in denser obstacle configurations it yielded a 36.98-metre route with only four turns and low computational overhead.

Autonomous navigation across modern industrial environments, including automated warehouse facilities and smart manufacturing workshops, fundamentally relies on calculating collision-free trajectories from a defined starting position to a target destination. While traditional robotic route planning techniques frequently suffer from sampling randomness or become trapped inside local minimum regions, intelligent metaheuristic algorithms provide versatile alternatives because they do not rely on gradient information. The Sine Cosine Algorithm has gained traction due to its mathematical simplicity and minimal parameter requirements, utilizing trigonometric functions to navigate toward optimal solutions. Nevertheless, conventional Sine Cosine formulations frequently experience search stagnation and insufficient convergence precision during later iterations when confronting complex obstacle layouts. Conversely, the Beetle Antennae Search algorithm offers rapid convergence and efficient single-agent local intensive exploration, but its fixed step parameters and lack of population diversity can restrict stability. The authors recognized that uniting modified variations of both algorithms could resolve these individual deficiencies.

To establish an effective global search stage, the study introduced architectural enhancements to the standard Sine Cosine Algorithm, creating an improved variant known as ISCA. In standard implementations, individual candidate positions update using trigonometric functions modulated by a linearly decreasing convergence parameter, which often contracts too quickly in early iterations and traps agents prematurely. To overcome this limitation, the authors implemented a nonlinear exponential-cosine decay strategy governed by an empirical parameter set to 0.05. This formulation slows parameter reduction during early iterations, allowing individuals to explore expansive regions of the solution space with large search steps before rapidly decreasing toward zero to consolidate solutions. Furthermore, the researchers incorporated an adaptive inertia weight that smoothly transitions between an upper bound of 0.9 and a lower bound of 0.4. Synchronizing this inertia weight with the nonlinear convergence factor maintains population diversity, prevents premature stagnation, and provides high-quality candidate routes for subsequent refinement.

Following the global exploration phase, an improved Beetle Antennae Search mechanism, designated as IBAS, conducts fine-grained local trajectory adjustments around the discovered route coordinates. Modeled after the foraging behaviors of longhorn beetles, the base algorithm evaluates fitness values sensed by simulated left and right antennae extending along normalized random directional vectors from the current position centroid. However, conventional beetle searches use a single agent and fixed step sizes, causing excessive stochastic wandering and unstable convergence. The proposed IBAS design resolves these limitations by introducing a nonlinear dynamic step-size decay strategy that systematically decreases movement increments across iterations while enforcing a minimum threshold of 0.1 to avoid search stagnation. Additionally, the researchers replaced the solitary searching agent with a multi-beetle parallel search strategy employing a subpopulation of ten beetles. By deploying multiple sensing agents simultaneously at the same route point and selecting the fittest candidate, the local optimization mechanism suppresses erratic movement and naturally yields smoother paths.

The architecture of the IBAS-ISCA framework relies on a serial cooperative optimization structure that enforces temporal separation rather than parallel execution. The researchers posited that global exploration and local exploitation operate most effectively when executed sequentially: the global search engine first identifies the broad geographic basin of optimal routes, after which the local refinement module systematically polishes the candidate coordinates. Before initiating iterative loops, the framework utilizes Opposition-Based Learning to optimize the initial population. This technique generates opposite solutions across coordinate boundaries and evaluates both candidate sets, selecting only the highest-performing individuals to populate the initial pool. Within each iteration cycle, route control points are first updated by the ISCA global exploration operators and subsequently transferred directly into the IBAS local refinement channel. To ensure that local stochastic searching never degrades established route quality, the authors integrated a greedy acceptance mechanism that compares fitness values before and after local refinement, strictly retaining the superior positional update.

Accurate mathematical representation of the physical environment and trajectory geometry forms the foundation of the proposed navigation system. The operational workspace is represented using a two-dimensional grid map with a spatial resolution of one metre, where coordinates represent continuous physical positions. Rather than optimizing every intermediate coordinate, which would introduce severe computational redundancy and high-dimensional complexity, the framework adopts cubic spline interpolation to fit a continuous, smooth trajectory across a sparse sequence of control points. Specifically, the model employs five control points for twenty-by-twenty maps and eight control points for fifty-by-fifty maps, with boundary clipping preventing coordinate excursion beyond workspace perimeters. A scalarized multi-objective fitness function evaluates candidate paths by combining total Euclidean length with weighted penalty terms for obstacle collisions and proximity safety. A binary collision penalty with a weight coefficient of 10.0 penalizes intersecting obstacles, while a continuous safety penalty with a weight coefficient of 5.0 penalizes trajectories encroaching within a predefined clearance of 0.5 metres.

The researchers verified the optimization capabilities of IBAS-ISCA by executing benchmark simulations in MATLAB R2021a on a workstation equipped with an Intel Core i7-12700H processor and 16 gigabytes of memory. Using a standardized population size of fifty individuals across five hundred iterations, the algorithm was tested against three contemporary optimizers: the Sparrow Search Algorithm-Grey Wolf Optimizer, the Black-winged Kite Algorithm, and the Modified Subtraction-Average-Based Optimizer. On both the unimodal Sphere function and the multimodal Rastrigin function, the hybrid model achieved rapid convergence, reaching logarithmic fitness values of negative 1.62 and negative 1.22, respectively, without exhibiting the stagnation observed in competing methods. Across thirty independent stochastic runs, Wilcoxon rank-sum tests confirmed statistically significant performance improvements with p-values below 0.05. Further evaluations across additional complex landscapes—including Griewank, Ackley, Rosenbrock, and deceptive Schwefel functions—demonstrated consistent superiority, with the hybrid model reaching a score of negative 11,763 on the Schwefel benchmark, which was closest to the global optimum among all tested algorithms.

To isolate and quantify the specific performance contributions of each algorithmic component, the authors conducted comprehensive ablation experiments on both the theoretical Sphere benchmark and a simulated twenty-by-twenty grid environment. The investigation examined four configurations: the complete IBAS-ISCA system, a variant lacking Opposition-Based Learning initialization, a variant omitting the improved Sine Cosine global search modifications, and a variant operating without the improved Beetle Antennae local optimization module. In the benchmark tests, omitting Opposition-Based Learning noticeably slowed convergence during early iterations, confirming its value in establishing high-quality starting populations. Removing the ISCA modifications triggered search stagnation after approximately one hundred and fifty iterations, while removing the IBAS module degraded later-stage convergence precision. In the simulated routing task, removing ISCA produced a 16.02 percent increase in total path length and increased the turn count from three to seven, confirming its critical role in global distance minimization. Removing IBAS increased the cumulative turning angle by 66.15 percent, illustrating its decisive role in smoothing trajectories.

Practical route planning performance was assessed across thirty independent trials using identical sets of thirty distinct random seeds to ensure rigorous paired comparisons under a fixed budget of nine thousand function evaluations. In a twenty-by-twenty simple grid mimicking warehouse shelving, IBAS-ISCA planned smooth paths averaging 29.22 metres with three turns and a total turning angle of 86.37 degrees, consuming 1.48 seconds of computation time. In a dense manufacturing layout, the hybrid model generated paths averaging 36.98 metres with four turns and an average runtime of 2.25 seconds, significantly outperforming competitors in turning count and turning angle. When scaled up to a fifty-by-fifty workspace, IBAS-ISCA produced a trajectory measuring 78.45 metres in 6.85 seconds with twelve turns and a cumulative turning angle of 156.8 degrees. Classical A-star planning completed faster at 0.85 seconds but lacked trajectory smoothness, whereas an extended seven-second matched-time evaluation of Rapidly-exploring Random Tree star yielded a longer 82.74-metre route and an extensive cumulative turning angle of 326.4 degrees.

To confirm that planned geometric trajectories remain executable under physical robotic constraints, the researchers conducted dynamic route tracking simulations in Gazebo on the fifty-by-fifty map using a differential-drive robot model with a 0.5-metre footprint diameter, a maximum linear velocity of 0.3 metres per second, and an angular velocity of 0.5 radians per second. Guided by a pure-pursuit controller utilizing ground-truth odometry, the robot successfully completed navigation without collisions while achieving the shortest tracked distance and smallest tracking error among all evaluated planners. Despite these successful demonstrations, the authors explicitly emphasized important technical limitations inherent to the study. The current algorithmic framework was validated exclusively within static grid environments and does not incorporate moving obstacles, real-time reactive re-planning, or explicit kinematic constraints during the primary optimization phase. Furthermore, the scalarized objective function relies on fixed weighting factors rather than generating a full Pareto front of non-dominated trade-offs. The authors indicated that integrating dynamic obstacle handling and Pareto multi-objective optimization represents the primary direction for future research.

Subject of Research: Mobile robot route optimization using hybrid metaheuristic algorithms

Article Title: A hybrid IBAS-ISCA route optimization model for mobile robots in static environments

Article References: Wang, X. (2026). A hybrid IBAS-ISCA route optimization model for mobile robots in static environments. Discover Artificial Intelligence, 6(1), Article 1793. https://doi.org/10.1007/s44163-026-02644-8

Image Credits: AI Generated

DOI: 10.1007/s44163-026-02644-8

Keywords: mobile robotics, route optimization, sine cosine algorithm, beetle antennae search, path planning, obstacle avoidance, metaheuristic algorithms, cubic spline interpolation, hybrid, IBAS-ISCA, route, optimization