In biological and social systems, a “swarm” refers to a group of individual units that behave as a single intelligent entity. “Swarm behavior,” the collective result of the local interactions among the group members, exhibits what is called “swarm intelligence.” By identifying the design principles of such swarm intelligence, we may be able to create swarm robots that are highly adaptable, fault tolerant, and dimensionally flexible.

An interdisciplinary approach, including disciplines ranging from technology to biology to the mathematical sciences, for example, is used to elucidate the design principles of swarm intelligence. We believe that such knowledge will lead to transformations in the field of swarm robotics.

This special issue highlights 19 exciting papers, including 13 research papers, five review papers, and one letter. Some papers focus on understanding the mechanism of real swarm phenomena, while the other papers focus on designing intelligent swarm systems. The keywords of the papers are as follows.

• Swarm intelligence

• Interdisciplinary approach

• Decentralized control

• Swarm robot

• Collective behavior

We would like to express our gratitude to all authors and reviewers, and we hope that this special issue contributes to future research and development in swarm intelligence.

Swarm intelligence is intelligence produced by multiple agents interacting with each other according to a simple set of rules, resulting in a system-wide intelligence. Such intelligence is found in a wide range of biological and social systems, and attempts have been made to understand the underlying principles through analytical approaches by biologists and sociologists and synthetic approaches by mathematical scientists and engineers. On the other hand, there are also attempts to construct artificial swarm intelligence systems that are not necessarily based on real-world phenomena. This review describes recent interdisciplinary research on swarm intelligence and its future prospects.

Much attention has been provided to autonomous decentralized systems based on swarm intelligence algorithms in robotics because of their resistance to component failure and ability to adapt to new environments. During development, various types of collectively migrating cells contribute to tissue and organ formation and have provided useful models for studying swarm behaviors. In the adult brain under physiological conditions, collective cell migration is almost exclusively observed in the rostral migratory stream, where adult-born new neurons travel long distances in contiguous chain-like formation. After ischemic stroke, some new neurons migrate toward the lesion site. Studies show that the promotion of migration is critical for efficient neuronal rewiring in the post-stroke brain in rodents. The new neurons traverse to injured tissues that are not conducive to migration by forming small chains, clearing a path through glial cells, and interacting with blood vessels. Although processes involved in migratory behavior, including cytoskeletal dynamics, intercellular adhesion, and chain formation, have been separately investigated, the mechanisms underlying neuronal swarm behavior are unclear. Future studies should help further our understanding of swarm intelligence and advance the development of novel strategies for controlling neuronal migration to promote efficient functional repair and rewiring in various pathological conditions.

Honey bees are social insects that form colonies (hives), which often consist of more than 10,000 individuals. In a colony, bees allocate jobs (division of labor) and work cooperatively and intelligently to maintain the colony’s activity, such as nursing broods, cleaning, and guarding against enemies. Among worker bees, only forager bees collect food, and success in finding food directly influences colony survival. For more efficient foraging, honey bees share location information pertaining to profitable food sources through specific behavior called “waggle dances.” During such dances, the direction and distance from the hive to the food source are encoded as body movements. Other foragers follow the dancing bees and receive location information. Some of these bees then fly to the advertised location to find the food source. Some of these “recruited bees” subsequently dance to recruit new bees. This process is then repeated. Consequently, many foragers visit the food source, and a colony can rapidly and flexibly collect large amounts of food even in foraging environment that can suddenly change (e.g., flowers disappear or nectar flux increases/decreases). To achieve effective food collection through the waggle dance, the behavior of both the dancers and followers probably contains information for an implementation of “swarm intelligence.” In this review, we introduce the properties of dance behavior at the levels of dancers, followers, and colonies. We found that errors in waggle dance information play an important role in adaptive foraging in dynamically changing environments.

We investigated the chemotactic behaviors of the nematode Caenorhabditis elegans, whose individuals have only 302 neurons but might sense the density of other individuals. As an individual detects areas with high concentration of a target chemical, mimicking their behavior may improve the exploration efficiency of autonomous distributed agents with limited sensing area and no direct communication with others. Inspired by this behavior, we experimentally determined the relationship between the density of individuals and probability of rapid turns to develop a search algorithm. We found a parameter set of “elite” individuals that achieved a high similarity of individual distributions with respect to a chemical gradient. Then, we implemented a motion selection algorithm that reflects the observation results so that an autonomous distributed agent, which has limited sensing range, achieves effective searching in a multi-peak environment. We simulated autonomous agents and applied the parameter sets obtained from elite, inferior, and single individuals. Through verifications using various benchmark potential functions, we concluded that the parameters of the elite group improved the search efficiency.

A model is proposed for group chase and escape using chemotactic movements only. In the proposed model, the movement depends on the concentration of the chemical substances released by each agent. Chemotaxis-based interactions propagate slower and later, and exist locally between agents, making groups chase and escape under more uncertain circumstances than in cases where agent distance measurements use electromagnetic waves, such as visible light. Numerical results with the model demonstrate that maintaining a longer distance between the chasers and targets is a better strategy for each group.

This article provides comparative perspectives on collective behaviors that are widely found throughout the animal kingdom, ranging from insect and crustacea swarms, fish schools, bird flocks, and mammal herds to human crowds. Studies of nonhuman animal and human collective behaviors have progressed almost separately even though they have a similar history. Theoretical studies have investigated the reproduction of collective phenomena from simple inter-individual rules, and subsequent empirical and experimental studies have found diverse and complex collective behaviors that are difficult to explain with classical theoretical models. As a consequence, a wide variety of interaction rules have been proposed. To determine models to be implemented in nature and find fundamental mechanisms of collective behaviors, this paper argues that we should compare collective behaviors among various species while adopting Tinbergen’s four questions regarding mechanism, function, development, and evolution as a methodological basis. As an example of a comparative collective behavior paradigm, we introduce our studies in which a mutual anticipation mechanism inspired by nonhuman animal collective behaviors can be linked to a self-organization function in human collective behaviors. We expect that the study of comparative collective behaviors will expand, the methodology will become more sophisticated, and new perspectives regarding the multitemporal features of collective behaviors will emerge.

Coordinated movement of self-propelled agents has been well studied in collectives or swarms that display directional movement. Self-propelled agents also develop stable spatial patterns in which the agents do not necessarily exhibit directional collective movement. However, quantitative measures that are required to analyze the local and temporal coordinated movements during pattern formation processes have not been well established. Here, we study the coordinated movement of individual pairs of two different types of cells in a freely moving cell population. We introduced three criteria to evaluate coordinated movement in live imaging data obtained from the abdomen of the fruit fly, Drosophila melanogaster, at the pupal stage. All three criteria were able to reasonably identify coordinated movement. Our analysis indicates that the combined usage of these criteria can improve the evaluation of whether a pair of cells exhibits coordinated movement or not by excluding false positives. Quantitative approaches to identifying coordinated movement in a population of freely moving agents constitute a key foundational methodology to study pattern formations by self-propelled agents.

In recent years, cooperative transportation systems using multiple mobile robots have been studied. The advantage will be to transport objects that cannot be dealt with by a single robot and transport them by using smaller robots. Although cooperative transportation by a group of robots has been studied, the conventional transportation targets are limited to objects whose posture is stable. In this paper, we propose a system in which robots pile up on each other to support an object, aiming at a system for more versatile object transportation, including unstable objects. After deriving the conditions by modeling the support system in object transportation, we verify the transporting performance including the robotic pile-up effect through actual robot experiments.

In recent years, research has been conducted on swarm robot systems in which multiple autonomous mobile robots cooperate to perform tasks. Swarm robot systems are expected to perform high functionality as a group by cooperating with each other, in spite of the limited capabilities of the individual robots. This paper explores a method of simplifying swarm robot controllers as much as possible for swarm robot navigation. If we can achieve autonomous navigation of swarm robots to a target area with minimal resource consumption, they only need to implement the task execution function in that area. This leads to lower costs for swarm robot design and more efficient system architecture design. To address the above aims, this study focuses on heterogeneity. Specifically, we introduce a navigator robot that indirectly guides swarm robots named the worker robots. Heterogeneity in this paper refers to the worker robots and the navigator robots. We design the interaction between the navigator robot and the worker robots to provide a system that guides the worker robots to the destination.

This study considers a simple robot swarm navigation system based on shepherding in an environment with obstacles. Shepherding is a system in which a small number of control agents (shepherds and sheepdogs) indirectly guide several robots (sheep) by driving them from behind. Previous studies have predominantly focused on verifying proposed controllers based on numerical simulations and navigation experiments in well-prepared environments. However, additional shepherding experiments need to be conducted in environments with obstacles. This study aims to facilitate shepherding-type swarm robot navigation in an environment where a wall obstructs the goal. Usually, a high-end controller design is adopted for the robot to prevent it from getting trapped by obstacles. However, as the environment becomes more complex, the system design may become difficult. In contrast, this study proposes a simple shepherding navigation system based on creating and controlling “fields” to avoid obstacles. This research aims to verify whether the robot can be guided to a goal without obstacle recognition by creating an acoustic field based on the diffraction effects of sound. The proposed method modifies the previous shepherding models for sheep and shepherd robots to make them behave according to the acoustic field gradient. We demonstrate the validity of the proposed system by performing robot navigation for dog and sheep robots.

Swarm robotics can cooperatively perform large, multiple tasks by controlling a swarm composed of many robots. Currently, approaches for operating multiple robot swarms are being studied for further evolution of this system. This study addresses a multiple movement task in which robot swarms move collectively in the same environment. In this task, the movement paths of robot swarms must pass each other in a cooperative manner when they intersect. The robots in this system behave under autonomous distributed control, thus must consider a passing behavior suitable for their own situation. This study proposes a turning behavior based on the chaos theory to ensure that a robot swarm avoids other approaching robot swarms. Each robot swarm applying the proposed method passes other swarms while autonomously deciding its turning direction and continuing its own collective movement task. In addition, the decision making based on the chaos theory predicts future values according to the current value. Therefore, it is expected to be useful for task scheduling. The performance of multiple robots passing each other is evaluated with the proposed method using numerical simulations. This performance shows that the robot swarms can avoid each other without collision using the closest inter-robot distance as the evaluation metric. Finally, robot swarms with varying shapes and scales complete their own movements in an environment where these movement paths intersect at a single point.

This paper presents a method of generating collective behavior of a multi-legged robotic swarm using deep reinforcement learning. Most studies in swarm robotics have used mobile robots driven by wheels. These robots can operate only on relatively flat surfaces. In this study, a multi-legged robotic swarm was employed to generate collective behavior not only on a flat field but also on rough terrain fields. However, designing a controller for a multi-legged robotic swarm becomes a challenging problem because it has a large number of actuators than wheeled-mobile robots. This paper applied deep reinforcement learning to designing a controller. The proximal policy optimization (PPO) algorithm was utilized to train the robot controller. The controller was trained through the task that required robots to walk and form a line. The results of computer simulations showed that the PPO led to the successful design of controllers for a multi-legged robotic swarm in flat and rough terrains.

Embodied evolution is an evolutionary robotics approach that implements an evolutionary algorithm over a population of robots and evolves while the robots perform their tasks. In embodied evolution, robots send and receive genomes from their neighbors and generate an offspring genome from the exchanged genomes. This study focused on the effects of the communication range for exchanging genomes on the evolvability of embodied evolution. Experiments were conducted using computer simulations, where robot controllers were evolved during a two-target navigation task. The results of the experiments showed that the robotic swarm could achieve better performance by reducing the communication range for exchanging genomes.

This study focuses on mutation-based evolving artificial neural network (MBEANN), a topology and weight evolving artificial neural network (TWEANN) algorithm. TWEANN optimizes both the connection weights and neural network structure. Primarily, MBEANN uses only mutations to evolve artificial neural networks. An individual in an MBEANN is designed to have a set of sub-networks called operons. Operons are expected to have functions during evolution because they do not recombine with other operons. In this study, we applied MBEANN to design a controller for a robotic swarm on cooperative transport, where the following canonical evolving artificial neural network (EANN) methods do not work well. For comparison with MBEANN, we used an EANN with a fixed network structure and neuroevolution of augmenting topologies (NEAT), which is a widely used TWEANN algorithm. We confirmed that the robot controller that evolved with the MBEANN outperformed the structure-fixed EANN and NEAT controllers. In addition, we investigated the behavior of the swarm robot obtained using the proposed method, in which we deactivated each operon to extract its function. The results show that operons could have their functions, and that several operons could strengthen one another’s functions.

Swarm robotics system (SRS) is a type of artifact that employs multiple robots to work together in a coordinated way, inspired by the self-organizing behavior of social insects such as ants and bees. SRSs are known for their robustness, flexibility, and scalability. This study focuses on evolutionary robotics (ER) which uses artificial neural networks (ANNs) as controllers to operate autonomous robots. In traditional ER research, SRSs were often composed of teams of homogeneous robots, each of which is controlled by a single ANN. In contrast, this study focuses on the implementation of ER in a heterogeneous SRS. To evaluate our approach, we present the concept of employing multiple controllers for sub-teams in a swarm. Heterogeneity was achieved using different controllers for the same physical bodies. We simulated a cooperative transport task, in which the performance of heterogeneity was superior because the two ANN controllers were able to express a variety of behaviors as an entire swarm. Additionally, this study investigated how well the three types of parental selection methods of the heterogeneous approach, can help to optimize the performance of the swarm.

In this study, we added voting behavior in which voting proportionately reflects the value of a view (option, opinion, and so on) to the BRT agent. BRT agent is a consensus-building model of the decision-making process among a group of human, and is a framework that allows the expression of the collective behavior while maintaining dispersiveness, although it has been noted that it is unable to reach consensus by making use of experience. To resolve this issue, we propose the incorporation of a mechanism of voting at frequencies proportional to the value estimated using reinforcement learning. We conducted a series of computer-based experiments using the box-pushing problem and verified that the proposed method reached a consensus to arrive at solutions based on experience.

This study proposes a bi-connectivity control algorithm for a multi-robot system consisting of a robot that communicates through line-of-sight communication. Optical observation and communication require a line of sight to a subject, and a robot in a multi-robot system sometimes impedes the line of sight between other robots. The motion of individual robots may violate the connectivity of the communication network. A distributed control strategy for bi-connectivity was constructed by modifying a previous method that does not consider line-of-sight restrictions. The results of the numerical simulation and experiments with actual robots are reported to emphasize the validity of the proposed control method.

A McKibben-type pneumatic actuator (MPA) is a soft actuator that generates tension by inflating a rubber tube with compressed air. Electropneumatic regulators are typically employed to regulate air pressure in MPAs. However, they are normally large in size and expensive, which are significant obstacles to the autonomous decentralized control of many MPAs in achieving various robot motions. In this study, the exerted tension of the MPA was controlled using a small solenoid valve that could be opened and closed instead of an electropneumatic regulator. To achieve this tension control, we proposed the use of a dynamic quantizer that converts continuous pressure values into discrete pressure values and controls the solenoid valve based on the discretized pressure values. The proposed method was applied to feedforward and feedback control of the exerted MPA tension under isometric conditions. Experiments on an actual device with a small solenoid valve demonstrated the effectiveness of the proposed method based on a dynamic quantizer.

Biomolecular motor-based micro-sized robots have recently created an innovation in the field of science and technology as molecular transporters. Groups of these tiny robots can work substantially better than individual ones in terms of the transported distance and number or size of cargo. Site-specific molecular delivery, the main feature of these robots, has helped to improve the workability of robots in a more controllable manner.

Using fire blankets is one effective option for fire extinguishing. This paper proposes a two-step release mechanism for unmanned aerial vehicles (UAVs) that can precisely drop a blanket onto a target location using the advantage of wind load, avoiding the ground effect. We analyzed wind velocity under the UAV body to find the optimized range that maximizes the release of the fire blanket. Afterwards, we tested the performance of our proposed system in an indoor environment with simulated wind disturbance. Our experiment results showed that the system is robust against wind disturbance at 3.5 m/s. Finally, we tested our system with UAVs in an outdoor environment with different heights and proved the effectiveness of the system for up to 4 m height in an outdoor environment.

Recently, several indirectly controlled sliding robots have been designed to achieve efficient and stable locomotion on slippery surfaces, and numerical simulations proved their possibility. However, it is difficult to achieve the same performance on a real machine because the wobbling mass, composed of springs and dampers, shows unexpected behavior when it is moved in translation. In this study, we propose a model of an arc-shaped sliding locomotion robot with a rotating wobbling mass. Specifically, the position of the center of gravity of the robot was changed, and the pull-in phenomenon due to the rotational motion of the wobbling mass is utilized. By rotating the wobbling mass, the sliding motion is realized while maintaining a strong propulsive force. First, we performed experimental validation of the proposed new mechanism. Subsequently, a detailed mathematical model was constructed for numerical analysis. Finally, the motion performance was optimized by the Bayesian optimization method.

Many small ground vehicles that travel on sidewalks, such as electric wheelchairs, are equipped with a differential-wheeled mechanism to turn precisely on the spot and accurately control their trajectory. Conventional ground vehicles operating on public roads have action indicator devices such as blinkers and brake lights, which are effective in preventing collisions during operations. However, similar blinkers may not be suitable for sidewalk navigation because the differential-wheeled mechanism allows a neutral turn without considering the surroundings. In this study, we developed a new action-intention indicator for future differential-wheeled mobile robots that employs RGB light-emitting diode strip lights to express various colors and complex blinking patterns. Sensitivity evaluation experiments were conducted with approximately 50 subjects to determine the optimal colors and blinking patterns for a differential-wheeled mobile robot. Additionally, reaction-time acquisition experiments were conducted to confirm the effectiveness of the proposed intention-display device for mobile robots. Finally, we evaluated whether the determined blinking patterns exhibited appropriate power consumption.

There are many methodologies assisting in the detection and tracking of trapped victims in the context of disaster management. Disaster management in the aftermath of such sudden occurrences requires preparedness in terms of technology, availability, accessibility, perception, training, evaluation, and deployability. This can be achieved through intensive test, evaluation and comparison of different techniques that are alternative to each other, eventually covering each module of the technology used for the search and rescue operation. Intensive research and development by academia and industry have led to an increased robustness of deep learning techniques such as the use of convolutional neural networks, which has resulted in increased reliance of first responders on the unmanned aerial vehicle (UAV) technology equipped with state-of-the-art computers to process real-time sensory information from cameras and other sensors in quest of possibility of life. In this paper, we propose a method to implement simulated detection of life in the sudden onset of disasters with the help of a deep learning model, and simultaneously implement multi-robot coordination between the vehicles with the use of a suitable region-partitioning technique to further expedite the operation. A simulated test platform was developed with parameters resembling real-life disaster environments using the same sensors.

Recently, underwater robotics has rapidly developed, and is often used in open-water exploration and underwater operations in known environments. However, there are still several problems in exploring the interiors of complex underwater environments, which are essential for scientific exploration and industrial applications, such as caves and shipwrecks. This study aims to complete the exploration of the environment of structures under water bodies. A real-time manipulative small underwater robot was designed and developed. The robot’s autonomous depth control and linear motion-assisted control are also realized by real-time sensor data processing, which provides stability and operability to move in small areas and complex environments. The sonar system is used to construct a submap for small-area scanning. Finally, by combining the odometer algorithm and contour extraction, the submaps are stitched together to construct a complete map of the internal underwater environment.

The field of robotics involves extensive knowledge and skills related to numerous fields, and the World Robot Summit benchmarks technical skills in service robotics through competition. While grasping and moving a wide variety of objects, we examine a new method for handling objects using a suction mechanism. Notably, the proposed mechanism utilizes the adhesion force resulting from Stefan adhesion, which occurs when a solid plate detaches from a smooth surface. Experiments are conducted to evaluate the performance of the adhesion hand, and the obtained results demonstrate its effectiveness.