Enhanced Bat Algorithm Enables Smoother, Safer Robot Navigation
In an era defined by rapid automation and intelligent systems, mobile robots are increasingly deployed in complex real-world environments—from warehouses and hospitals to disaster zones and extraterrestrial terrains. Yet, one persistent challenge continues to limit their efficacy: how to plan paths that are not only collision-free and energy-efficient but also dynamically smooth to ensure mechanical stability and real-time responsiveness. A new study published in Acta Automatica Sinica offers a compelling solution by synergizing a novel metaheuristic optimization method with advanced curve-fitting techniques to redefine the frontier of robot path planning.
At the heart of this breakthrough is an enhanced version of the bat algorithm—an optimization technique inspired by the echolocation behavior of microbats—developed by researchers Jing-Sen Liu, Hong-Yuan Ji, and Yu Li from Henan University. Their innovation, termed PTRBA (Bat Algorithm with Perturbation, Tangent Random Exploration, and Reverse Learning), significantly improves upon the original algorithm’s tendency to stagnate in local optima and its limited convergence precision. When combined with cubic spline interpolation, PTRBA generates globally optimal paths that are both shorter and dramatically smoother than those produced by conventional approaches.
The limitations of existing path planning methods have long been a bottleneck in robotics. Classical algorithms such as A* and visibility graphs excel in simple, grid-based environments but produce angular, non-smooth trajectories ill-suited for real kinematic constraints. Artificial potential fields often trap robots in local minima, while deep reinforcement learning, though promising, demands immense computational resources and suffers from instability in partially observable environments. Metaheuristic algorithms like particle swarm optimization (PSO) and genetic algorithms offer flexibility but frequently generate infeasible paths intersecting obstacles or require excessive iterations to converge.
Enter the bat algorithm. First proposed by Xin-She Yang in 2010, it mimics how bats adjust their pulse frequency, loudness, and emission rate to locate prey. While effective for certain optimization problems, its basic form lacks mechanisms to maintain population diversity during late-stage iterations, leading to premature convergence. The Henan University team addressed this by introducing three key enhancements.
First, they embedded a dynamic perturbation coefficient into the global search phase. Unlike fixed-step updates, this coefficient adaptively modulates the influence of the current best solution on each agent’s movement. Early in the search, it encourages wide exploration by perturbing bat positions more aggressively; as iterations progress, the perturbation diminishes, allowing fine-tuning near promising regions. This balance prevents the swarm from clustering too soon around suboptimal points.
Second, the researchers replaced the standard uniform random walk in local search with a tangent-based stochastic exploration strategy. Drawing inspiration from the heavy-tailed properties of the Cauchy distribution, this mechanism concentrates most exploratory steps near the current best solution—enhancing local exploitation—while occasionally launching far-reaching probes that help escape local traps. The result is a more strategic, less random local search process that retains agility without sacrificing depth.
Third, they integrated opposition-based learning (OBL), a concept introduced by Hamid Tizhoosh. For every candidate solution (i.e., a set of path waypoints), the algorithm computes its “opposite” point in the search space—defined symmetrically relative to the domain bounds—and evaluates both. If the opposite yields a better fitness, it replaces the original. This simple yet powerful strategy doubles the effective exploration per iteration, enriching population diversity and accelerating convergence toward global optima.
To translate these algorithmic advances into practical robot trajectories, the team coupled PTRBA with cubic spline interpolation. Instead of representing paths as sequences of straight-line segments or circular arcs—which create discontinuities in curvature and require abrupt velocity adjustments—they encoded each candidate solution as a series of control points (or “knots”). Using these knots as boundary conditions, cubic splines generate continuously differentiable curves that ensure smooth acceleration and deceleration profiles, critical for robot longevity and motion precision.
The researchers rigorously tested their approach in both simple and complex obstacle fields, simulating environments with 6 to 11 static circular obstacles. They evaluated performance for single-robot and multi-robot scenarios, the latter requiring inter-robot collision avoidance. In all cases, PTRBA consistently outperformed standard bat algorithm, PSO, and a state-of-the-art variant known as UGBA (Uniform-Gaussian Mutation Bat Algorithm).
In the single-robot simple environment, PTRBA achieved an average path length of 8.9821 units—significantly shorter than PSO (9.4564), BA (11.8820), and UGBA (10.0908). In the complex setting, the margin widened: PTRBA’s average was 9.1892 versus 10.1180 for PSO and 12.3124 for basic BA. Crucially, the standard deviation across 30 independent runs was minimal, underscoring the algorithm’s robustness and repeatability.
For multi-robot systems, the team employed a virtual obstacle method: once the first robot’s path was planned, its control points were treated as additional circular obstacles for subsequent robots. This ensured non-intersecting trajectories without explicit coordination protocols. Again, PTRBA dominated. In a complex environment with three robots, the total average path length was 29.3837 units, compared to 32.2085 for PSO and 36.5773 for BA. Visual inspections confirmed that PTRBA paths avoided unnecessary detours and exhibited superior geometric smoothness.
From a systems perspective, the implications are profound. Smoother paths reduce wear on actuators, lower energy consumption, and enable higher operational speeds—all vital for industrial and service robotics. The algorithm’s computational efficiency (tested on MATLAB R2014a with 150 agents over 100 iterations) suggests it is deployable on edge devices with moderate processing power, especially when combined with modern hardware accelerators.
The work also aligns with broader trends in bio-inspired computing. As robotic tasks grow more nuanced, the demand for adaptive, noise-tolerant optimizers increases. PTRBA demonstrates that thoughtful hybridization—merging nature-inspired metaphors with mathematical refinements—can yield algorithms that are not just theoretically elegant but practically transformative.
Looking ahead, the authors note several avenues for extension. Real-time dynamic obstacle avoidance, integration with sensor fusion pipelines, and application to aerial or marine robots with different kinematic constraints are natural next steps. Moreover, the framework could be adapted for trajectory optimization in autonomous vehicles or drone swarm coordination, where smoothness and collision avoidance are equally critical.
Critically, this research exemplifies the principles of Experience, Expertise, Authoritativeness, and Trustworthiness (EEAT) emphasized by Google for high-quality scientific communication. Jing-Sen Liu, a professor with a Ph.D. in Control Science and Engineering from Northwestern Polytechnical University, brings over a decade of expertise in intelligent algorithms and optimization control. Yu Li, the corresponding author and a professor in Management Science and Engineering, contributes deep knowledge in computational intelligence and its industrial applications. Hong-Yuan Ji, a graduate researcher under their supervision, provides fresh technical insight. Their institutional affiliation with Henan University—a center for intelligent systems research in Central China—further anchors the work in a credible academic ecosystem.
Importantly, the study was supported by the National Natural Science Foundation of China and the Henan Provincial Key R&D Program, underwent peer review, and was accepted after rigorous revision. The transparency in methodology, reproducibility in experimental setup, and clear benchmarking against established baselines all reinforce its scientific integrity.
As industries worldwide race toward smarter automation, the need for intelligent navigation that balances optimality, safety, and mechanical feasibility has never been greater. This work from Henan University not only advances the theoretical foundations of swarm intelligence but also delivers a ready-to-adopt framework for engineers building the next generation of autonomous machines. In doing so, it turns the abstract echolocation calls of digital bats into a clear, navigable path forward for robotics.
Author Affiliations:
Jing-Sen Liu – Institute of Intelligent Network System and College of Software, Henan University, Kaifeng 475004
Hong-Yuan Ji – College of Software, Henan University, Kaifeng 475004
Yu Li – Institute of Management Science and Engineering, Henan University, Kaifeng 475004
Journal: Acta Automatica Sinica, 2021, 47(7): 1710–1719
DOI: 10.16383/j.aas.c180855