New Algorithm Enables Robot Swarms to Form and Navigate in Limited Communication Environments
In a significant advancement for swarm robotics, researchers from South China University of Technology have developed a novel distributed control algorithm that allows large groups of mobile robots to form precise linear formations and navigate complex environments—even when communication between units is severely restricted. The breakthrough, published in the peer-reviewed journal Control Theory & Applications, addresses one of the most persistent challenges in multi-robot systems: how to achieve coordinated group behavior without relying on centralized control or constant, high-bandwidth communication.
The research, led by Associate Professor Jiaxiang Lou and senior engineer Haiming Liu, introduces a behavior-based approach that enables robots to make intelligent, autonomous decisions using only local sensor data and information from immediate neighbors. Unlike traditional formation control methods that depend on a designated leader or global knowledge of the swarm’s configuration, this new algorithm empowers each robot to dynamically adjust its movement based on real-time environmental feedback and predictive modeling.
As robotic swarms are increasingly deployed in real-world applications—from search and rescue missions in disaster zones to coordinated agricultural monitoring and warehouse logistics—the limitations of communication become a critical bottleneck. In GPS-denied environments such as underground tunnels, dense forests, or urban canyons, maintaining continuous, high-fidelity communication among hundreds or thousands of robots is not only impractical but often impossible. Existing formation algorithms, which assume robust network connectivity, tend to fail under such constraints, leading to disorganized swarms, inefficient movement, or even collisions.
The team’s solution, detailed in their July 2021 paper, circumvents these issues by combining elements of decentralized control, local interaction rules, and adaptive behavior switching. At the core of the algorithm is a hybrid strategy that blends the strengths of two well-known but previously separate approaches: “one-side following” and “two-side following” behaviors. In one-side following, a robot adjusts its position based solely on a single neighboring unit, typically the one directly ahead or behind in the desired formation. While this method can lead to rapid alignment, it often results in oscillatory behavior and high energy consumption, as each robot reacts to the movements of another that is itself still adjusting.
In contrast, two-side following involves a robot referencing both its nearest predecessor and successor in the formation, aiming to position itself at the midpoint between them. This bilateral approach promotes stability and reduces energy expenditure but can be slow to converge to the desired linear shape, especially in large swarms where initial positions are highly scattered.
The innovation from the South China University of Technology team lies in their dynamic switching mechanism. Robots begin the formation process using the two-side following strategy, allowing the swarm to quickly coalesce into a rough linear configuration. Once the group approaches the desired alignment, individual robots use a prediction model based on their recent velocity and the relative positions of their neighbors to determine whether they are slightly ahead, behind, or off to the side of their ideal position. This prediction enables them to selectively switch to a one-side following mode, targeting the most beneficial neighbor for fine-tuning their placement. This hybrid approach achieves the best of both worlds: fast initial convergence and precise final alignment, all while minimizing unnecessary movement and energy use.
“This algorithm is designed for scalability and robustness,” explained Haiming Liu, the corresponding author of the study. “In large-scale deployments, you can’t rely on every robot knowing the state of every other robot. Our method ensures that even with minimal communication, the swarm can self-organize, maintain formation, and adapt to obstacles in real time.”
The algorithm also incorporates sophisticated obstacle avoidance and collision prevention mechanisms. When a robot detects an obstacle within a predefined “avoidance distance,” it generates a lateral velocity component perpendicular to the swarm’s intended direction of travel. This allows the formation to split temporarily, navigate around the obstacle, and then seamlessly reassemble on the other side. The decision to activate avoidance behavior is based on the relative angle of the obstacle with respect to the swarm’s movement direction, ensuring that robots do not waste energy avoiding objects that are behind them or otherwise not in their path.
To prevent collisions between robots, the system employs a repulsive velocity field that activates when two robots come within a “safety distance.” This field pushes the robots apart, ensuring safe separation without disrupting the overall formation. These safety behaviors are prioritized over formation control, meaning that if a collision risk is detected, the robots will prioritize avoiding the crash even if it means temporarily breaking rank.
The researchers validated their approach through extensive simulations involving up to 200 robots in both obstacle-free and cluttered environments. In one scenario, a group of eight robots started from random positions and successfully formed a linear formation while navigating around three circular obstacles. The robots dynamically split and reformed the line, demonstrating the algorithm’s ability to handle external disturbances and return to a stable state—a hallmark of a robust control system.
In larger-scale tests with 50 to 200 robots, the team compared their hybrid algorithm against both pure one-side and pure two-side following methods. The results were compelling. The new algorithm achieved formation convergence significantly faster than the two-side method, often reaching the target configuration in less than half the time. At the same time, it consumed substantially less energy than the one-side method, reducing total energy expenditure by up to 40% in some cases. This efficiency is crucial for real-world applications where battery life is a limiting factor.
The performance gains stem from the algorithm’s intelligent use of local information. By predicting their position within the emerging formation, robots can make more informed decisions about which neighbor to follow, reducing the “chasing” behavior that plagues pure one-side strategies. The prediction is based on a weighted average of the robot’s recent velocity, filtered to smooth out transient fluctuations. This allows the system to distinguish between deliberate course corrections and noise, leading to smoother and more stable convergence.
Another key advantage of the approach is its complete decentralization. There is no single point of failure. If one robot fails or loses communication, the rest of the swarm can continue to function without interruption. This resilience makes the system ideal for missions in unpredictable or hostile environments where reliability is paramount.
The implications of this research extend beyond academic interest. In industrial automation, swarms of robots could coordinate to transport materials in a factory, dynamically adjusting their formation to navigate narrow aisles or avoid moving machinery. In agriculture, drone swarms could fly in formation to survey crops, dispersing to cover more ground when needed and re-forming for efficient return flights. In defense and security, unmanned aerial or ground vehicles could conduct coordinated surveillance, maintaining formation while evading detection or navigating complex terrain.
The work also contributes to a broader shift in robotics toward bio-inspired, self-organizing systems. Just as flocks of birds or schools of fish achieve complex group behaviors through simple local rules, this algorithm demonstrates how sophisticated coordination can emerge from decentralized, rule-based interactions. It reflects a growing understanding that sometimes, less is more—less communication, less central control, and more local intelligence can lead to more robust and scalable systems.
The publication of this research in Control Theory & Applications, a respected journal in the field of systems and control engineering, underscores its technical rigor and significance. The team’s stability analysis, based on Lyapunov theory, provides a solid mathematical foundation for the algorithm’s performance, proving that the system will converge to the desired formation under the specified conditions. This level of theoretical validation is essential for gaining acceptance in both academic and industrial circles.
While the current work is based on simulation, the researchers are optimistic about its real-world applicability. The algorithm’s reliance on standard sensor inputs—such as relative position and velocity—means it can be implemented on existing robotic platforms with minimal modifications. Future work will focus on physical experiments with robot swarms, testing the algorithm under real-world conditions with sensor noise, communication delays, and mechanical imperfections.
The team also plans to extend the algorithm to more complex formation shapes beyond straight lines, such as circles or grids, and to incorporate dynamic reconfiguration, allowing the swarm to change its shape on the fly in response to mission requirements. Additionally, they are exploring ways to integrate machine learning techniques to allow the swarm to adapt its behavior based on experience, further enhancing its autonomy and flexibility.
The success of this project is a testament to the growing strength of robotics research in China. South China University of Technology, located in Guangzhou, has established itself as a leading institution in automation and intelligent systems. The research was supported by grants from the Guangdong Provincial Department of Science and Technology and the Fundamental Research Funds for Central Universities, highlighting the importance placed on advancing technological innovation at the national level.
For engineers and researchers working in multi-robot systems, this algorithm offers a powerful new tool for designing scalable, resilient, and energy-efficient swarms. It represents a significant step forward in the quest to create truly autonomous robotic collectives that can operate effectively in the messy, unpredictable real world.
As robotic swarms move from the lab to the field, the ability to function under communication constraints will be a defining factor in their success. The work of Lou, Liu, and their colleagues provides a clear path forward, demonstrating that with the right algorithms, even the most challenging environments can be navigated by intelligent, coordinated teams of robots.
The implications are profound. In the not-too-distant future, we may see swarms of robots working together in disaster zones, where communication is spotty and conditions are chaotic, to locate survivors and deliver aid. We may see fleets of autonomous vehicles coordinating on highways without the need for constant cloud connectivity. We may even see space exploration missions where swarms of small satellites form and re-form in orbit, conducting observations with unprecedented flexibility.
This research is not just about robots forming lines. It’s about creating systems that are robust, adaptive, and capable of intelligent collective behavior under real-world constraints. It’s about building the foundation for a future where machines can work together as seamlessly and effectively as the natural systems that inspired them.
In a world increasingly defined by interconnected technologies, the ability to function when those connections are weak or broken may be the most important capability of all. The algorithm developed by the team at South China University of Technology is a major step toward that future.
Jiaxiang Lou, Zhenfeng Guan, Haiming Liu, He Cai, Huanli Gao, and Jie Huang, South China University of Technology, Control Theory & Applications, DOI: 10.7641/CTA.2021.00716