ABSTRACT. This paper focuses on tracking to the serial movement of the optimal solution at a very short frequency in dynamic environments. The dynamic environment is an environment in which the optimal solution changes over time. The optimization problem with such the environments is called Dynamic Optimization Problems (DOPs). To tracking to such a change, it is important to tracking the individuals at the edge of group. This is because they tend to be the best individuals when dynamic changes occur in unimodal problems. This paper proposes the mechanism that individual can search with jumping over the current best individual or other individuals that based on such a trend. This mechanism can be applied to various swarm intelligence algorithms. To investigate the applicability to swarm intelligence algorithms and tracking ability to continuous change of this mechanism, two algorithms (PSO and SSO) with two improvement incorporating this mechanism were applied to unimodal DOPs. The experimental results revealed that the following implications: (1) JO mechanism was effective in PSO; (2) FB mechanism is especially effective in PSO; (3) Combination of JO and FB was most effective in SSO.

ABSTRACT. Supervisory control of swarms is essential to their deployment in real world scenarios to both monitor their operation and provide guidance. We explore mechanisms by which humans might provide supervisory control to swarms and improve their performance. Rather than have humans guess the correct form of supervisory control, we use artificial evolution to learn effective human-readable strategies to help human operators understanding and performance when controlling swarms. These strategies can be thoroughly tested and can provide knowledge to be used in the future in a variety of scenarios. Behaviour trees are applied to represent human readable decision strategies which are produced through evolution. We investigate a simulated set of scenarios where a swarm of robots have to explore varying environments and reach sets of objectives. Effective supervisory control strategies are evolved to explore each environment using different local swarm behaviours. The evolved behaviour trees are animated alongside swarm simulations to enable clear understanding of the supervisory strategies. We conclude by identifying the strengths in accelerated testing and the benefits of this approach for scenario exploration and training of human operators.

ABSTRACT. Without any centralized control, the swarm robotic system exhibits considerable advantages when faced with individual level failure or the lack of global information. It has various applications, such as finding targets or item transportation in a relative dynamic environment. Spatial information has always been a necessity in generating collective transport behavior. Instead of global position information, which is a limitation for the application scenario, the visual information could be another input method for the robotic swarm. Benefit from the deep neural network technology, swarm robotic could extract spatial information from high dimensional information such as raw camera images captured by themselves. In this paper, the use of deep Q-Learning algorithm and several improved versions of it in training a robotic swarm to generate the collective foraging behavior is shown. The foraging behavior is evaluated in a transportation task, where the robots have to forage foods to a specific location while avoiding obstacles within limited time steps. Simulation results show that the proposed approach can generate collective foraging behavior from high dimensional raw camera image inputs for robotic swarms.

ABSTRACT. There are two types of ant society: one is monodomy in which a colony has a single nest as the nursery, and the other is polydomy that means the possession of multiple nests by a single colony. This difference is observed mostly at the interspecific level but sometimes within species. Why does such variation exist? We tested the hypothesis that this variation is related to the difference in colony-level strategy to counter environmental heterogeneity in resource distribution. We compared responses to heterogeneous resource distribution between monodomous and polydomous ants by focusing on the genus Pheidole (monodomous P. noda vs. polydomous P. megacephala). They are both polygynous and similar in worker body size. First, we tested the physiological integration hypothesis that maintains that nests in a polydomous colony exchange complementary resources, thereby the colony as a whole is physiologically integrated. Second, using the polydomus ant, we tested the hypothesis that nests are not only sharing complementary resources, but each nest tends to specialize in collecting one resource under spatially heterogeneous environments (nest-level division of labor). A series of laboratory experiments with nests connected by tubes revealed that when complementary foods (lacking protein or lacking carbohydrates) were provided to each of the connected nests, the polydomous ant flourished. However, when nests were disconnected by plugging the tubes, the polydomous ant almost totally failed, supporting the physiological integration hypothesis. In marked contrast, the monodomous ant kept the highest performance over five weeks even when only a nutritionally biased food was provided. This suggests that they store a large amount of nutrients in adult bodies that can be used when the outside food availability becomes poorer. The above suggests that polydomous ants and monodomous ants might adopt different strategies to heterogeneity in resource distribution. Polydomy might be a strategy to counter it spatially by extending the area of resource searching, whereas monodomous ants might deal with it temporarily by withstanding resource depressed periods of time. We found that in the polydomous ant each nest tended to specialize in collecting one resource of which availability is locally high. Interestingly, under such nest-level division of labor the foraging activity of the entire colony was enhanced.

ABSTRACT. Information is crucial to effective decision making, but too much information can be as detrimental as too little. An overload of information can lead to delayed and/or less accurate decision making, manifested in the ‘burden of choice’ increasingly faced by humans. Groups can out-perform individuals in decision making, an effect known as the ‘wisdom of the crowds’. This benefit can be attributed in part to the fact that groups can combine the knowledge of different group-members via information pooling, enabling decisions to be based on a greater collective knowledge pool. However, information pooling could also increase the risk of information overload. If the increase in available information from information pooling is not complemented by a corresponding increase in information processing capacity, this could lead to a collective form of the burden of choice. Here, I present findings from an agent-based model developed to reproduce the dynamics of the collective decision-making process during nest-site selection in a mass-recruiting ant species. In this model, individual agents scout an arena containing a variable number of candidate nests, and must collectively decide on which will be their new home. I use this model to explore the influence of group-size on information accumulation, and how this affects the speed and accuracy of collective decisions. I find that larger groups are more effective at sampling the environment that this this could render them more susceptible to information overload in candidate-rich environments. In these conditions, larger groups suffered delays to decision making because of poorly focussed information sharing. However, the cost of excessive information could be ameliorated if individuals are ‘fussy’ and filter information by withholding information on poor candidates. These results suggest groups face a potential collective burden of choice from excessive novel information in complex environments. However, collective information-filtering mechanisms analogous to those in the brains of unitary organisms could allow groups to exploit the wisdom of the crowds without suffering from information overload.

ABSTRACT. Social animals are able to collectively choose the most suitable resource among several options, maintaining in this way the cohesion of the group and the subsequent social benefits. These decisions, which can be critical for the survival of the group, can be also modulated by different factors, such as the individual physiological state, environmental conditions or group size.

Studies on gregarious and social arthropods have shown that the presence of conspecifics is able to amplify individual preferences through positive social interactions, leading to a better discrimination between the quality of resources. This classical approach of collective decision-making often underestimates individual complexity though.

The American cockroach (Periplaneta americana) is a model organism for many studies in biology, including research on aggregation and decision-making. We used a combination of experiments and computation modelling to study the mechanisms driving collective dynamics during the aggregation process of P. americana. First of all, we have demonstrated that these cockroaches show high rates of individual and group repeatability of behavioural traits over time (personality variation). Second, we show that significant differences observed among groups in terms of collective dynamics during the aggregation process depend on the variation of personalities inside the group. These results bring new insights regarding the influence of personality variation on decision-making, where key-individuals have an asymmetrical influence during the aggregation process and affect the exploitation and exploration of environmental resources by the entire group. Finally, we discuss the idea of potential regulation mechanisms of behaviour within groups when these key individuals are removed and other individuals must take their role.

ABSTRACT. Crab-like robots moved by a vibrating motor were given to soldier crabs (Mictyrus guinotae) in a laboratory condition, and their behaviors were recorded and analyzed. We conducted 1 crab and 5 crabs condition under 4 robots, compared with the control (no robot). Only in 5 crab condition, soldier crabs approached to the robots, and showed Levy walk move pattern which could be considered as the attribute of sociality. This result suggests that the soldier crab regarded the crab-like robots as swarm mates.

ABSTRACT. In urban transit network design problem (UTNDP), optimal route generation method has been studied from earlier, and most of these method focuses on improving passenger travel time and total routes length. However, conventional generated routes are not considered to adapt to environmental change, such as earthquake. Towards adaptation to environmental change, this paper proposes the alternative plans. The alternative plans only switch the bus stops to pass from basis transit route network so as to maintain the shape of transit route network. Also, these plans can adapt to environmental change such as decrease of total fleet size, collapsed situation of bus stops, and road closed due to disaster. Furthermore, by having alternative plans, the bus operator can choose the plan from various options at the disaster situation without the change of network shape. In the process of generating alternative plans, this paper employs the traffic volume on generated transit route network and difference of fitness value to determine where to pass. The difference of fitness value is between basis network and after switched network. Experiments on Mandl’s Swiss network problem has revealed the following implications: the proposed method (1) could generate various alternatives plans that spread across the solution space; (2) could generate an alternative plan that could cope with the abolition (road closed) of 2 routes from 6 routes; (3) succeeded in generating alternative plans with averagely 1.4% better performance to the basis transit network.

ABSTRACT. To provide a solution to the growing demand for secure controllers and to overcome the shortcoming of the previous research, we introduce a method to structure a tamper-resistant controller for a time-varying quantized control system in this paper. By utilizing piecewise approximation linearity of the trained neural network; a neural network, several signal processing modules and a time-varying quantizer are employed to realize the proposed controller approximately. At last, the feasibility of the proposed method is tested by a numerical example.

ABSTRACT. Our chloroplastic robots are able to passively move toward a light source in a reactive manner. In this paper, we apply the swarm robotic system to a task that requires the chloroplastic robots to cover the entire environment including shadow areas. In order for the robots to achieve the covering task, we propose a global path of the light source. The light source moving along the path illuminates shadow areas. Thus the robots are able to cover the entire environment. The global path is generated in consideration of the formation of robots performing the task. Through simulation experiments, the proposed path is compared to others generated for the traveling salesman problem, TSP, and evaluated in terms of the task efficiency and performance. Finally, we discuss the effectiveness of the global path for the covering task.

ABSTRACT. Individual ants in a colony can do suitable tasks based on their local situations without any directions from a leader. As a result of the individual behaviors, each ant has a specific task, which is so-called task allocation. The task allocation is a kind of self-organizing phenomena because any leader does not exist in a colony. Many researchers have studied the mechanism of the task allocation experimentally and theoretically. Ecologists [1] have proposed several hypothetical mechanisms based on their observations; a mechanism based on the age, a mechanism based on body size, and so on. As a theoretical model, Bonabeau et al. [2] proposed the response threshold (RT) model to describe the mechanism, which is the most important theoretical work nowadays. In experimental studies, quantitative observation methods have been established recently. Mersch et al. [3] observed the individual movement in the nest and the foraging field and analyzed the interaction among individuals using AR tags and high-frequency infrared camera. They quantitatively confirmed the proposed task allocation mechanisms. Despite the development of experimental quantitative analysis, the quantitative results have not been fed back into the theoretical researches. Yamanaka et al. [4] have shown that theoretical projections based on the RT model conflict with the long term observation results. In this study, we observed ants’ activity in an artificial nest and analyzed the time course data for long term observation using radio frequency identity (RFID) tags and sensors. We applied point process analysis to the obtained data and characterized the ants' activity data as a stochastic process. We also compare the characteristics of the stochastic process model with the reaction threshold model and discuss what kind of model modification is necessary.

ABSTRACT. In the last fifteen years, research on animal models has provided advances on how gut symbiotic microbes affect behavior and its underlying neurophysiology in animals, including humans. However, most studies on the gut microbiota only take into exam individual behavior without considering social dynamics. Contrarily, animals and humans live in complex societies where they constantly adjust physiology and behavior to social interactions. Therefore, to improve our understanding of how microbes and hosts interact and produce functional individual, social and collective phenotypes, we need to open our experimental approach to the group-level dimension. The ideal models for this purpose are social animals living in stable symbioses with microbes, such as eusocial insects. In our research, we investigate Camponotus carpenter ants and their obligate bacterial symbiont Blochmannia from a behavioral ecology perspective. By suppressing Blochmannia in single individuals using antibiotics, we aim to create ant colonies including differential proportions of bacteria-free individuals; then, using a machine learning-based video tracking system, we plan to study network features and group-level behavior of such experimental colonies.

ABSTRACT. Animals are well known to use their spatial memory to search or relocate their goals. They sometimes present self-avoiding walks in which agents do not revisit any previously visited locations and at other time they present self-attracting walks in which they make home-range behaviors. Interestingly, some animals seem to change their movement strategies drastically. In that sense, an artificial agent model illustrating a specific strategy does not always demonstrate animal movements adequately. Here, we developed an artificial agent model where the agent considered the continuity and the discontinuity of its memorized locations. In our proposed model, the agent sometimes regards part of its memorized cells as very past information and change its directional rule not to revisit those locations. Our results show that the agent alternates super-diffusive walks and sub-diffusive walks. Interestingly, the agent sometimes persists in self-avoiding (attracting) walks for a long period. These characteristic movements emerge without any adjustments of parameters. Moreover, we compared simulation data with experimental data of foraging ants, which was obtained using ANTAM and found that our model partially reproduced walking patterns of individual ants.

ABSTRACT. Among ants that show complex and diverse collective actions, army ants are known to raid prey in groups. It has been confirmed that the characteristic pattern changes from radial to tree-like with time in the swarm raid on the army ants. There are some simulation studies focusing only on the emergence of tree-like pattern. However, the models adopted in their simulations cannot represent the transition of patterns from radial to tree-like observed in the real world. In this study, we propose an army ant model with modification of the model adopted by Sole ́ et al. and clarify the conditions of the mergence of the transition. As a result of simulation experiments using our model, we found that simply modifying Sole ́’s model so that it can move in eight directions is not enough to express multidirectional to unidirectional convergence. In addition, our simulation experiments showed that in order to express the pattern transition from radial to tree-like, it is necessary for the individual to continue to move forward until get a food, and for the returning individual to completely follow the pheromone.

ABSTRACT. This paper proposes a new response threshold model based on the contact stimulus with the foraging ants in order to make the autonomous role allocation mechanism based on the local information for multi-agent system. In addition, we show that multi-agent system can optimize role allocation based on local information through computational experimentations.

ABSTRACT. In current multi-agent systems, there is a need for human operators to be able to communicate with agents in the field in real-time that goes beyond single agent teleoperation or executing pre-written programs. This includes being able to control the shape and movement of the system in a dynamically-shifting environment using an intuitive interface. While previous research has focused on moving the collective of agents as a whole, in this ongoing work, we establish that a swarm can be shaped to different formations via gesture control using a wearable (smartwatch) interface. This high-level command assists the swarm in completing its task by taking commands from a higher level operator/user.

ABSTRACT. The purpose of this work is to introduce a lighter-than-air robotics platform that is safe, easier to deploy with fewer operators, and can be used to test swarm behaviors and emergent behaviors on real agents. This work shows the results of having 25+ lighter-than-air autonomous agents (titled LTA3) fly simultaneously under a mixture of local and global control. Physical agents are used in place of simulated agents in order to ensure that all dynamics and noisy sensor feedback characteristics as well as unpredictable environmental errors are not lost. The results show a promising path forward for the ability to study emergent behavior with real-world agents.

ABSTRACT. Many of the swarms demonstrated today lack a key piece in order to be useful in the real-world: communication. Our most basic and commonly used communication infrastructure such as cellular networks and WiFi, assume that communication is facilitated by a centralized actor or controller. This falls short in a swarm setting. We demonstrate that the technology exists to take swarms to the next level and enable new fully decentralized swarms for novel applications.

ABSTRACT. In recent years, the use of drones in harsh environments has attracted attention. There is an expectation that human's complex work can be substituted if limited information processing ability and mechanical energy of a drone can be integrated by forming a swarm by making use of the high mobility of the drone. On the other hand, there are still many issues to achieve this. Especially in situations where human instruction cannot be obtained, the drone group needs to judge the situation on the site and take appropriate coordinated action. In this paper, it is shown that disaster response can be dealt by drones that perform tasks continuously while changing the three priorities of patrol, firefighting, and supply depending on the energy status of the drone. The conventional BRT model lacks a part to estimate the evaluation of options, and it has not been easy to use for practical problems as discussed here.

ABSTRACT. Biological swarms - from the simplest bacteria to large populations - compute the solution of geometrical problems for their growth and survival. In order to engineer synthetic swarms, it is of interest to understand how biological swarms relate space and time to sense their geometry. Here we propose a swarm architecture that senses the shape of its environment, using agents that form Turing patterns as "sonar", and neural-like layers to process those Turing patterns and find the shape of reactor that generate them

ABSTRACT. Possibility theory is proposed as an uncertainty representation framework for distributed learning in multi-agent systems and robot swarms. In particular, we investigate its application to the best-of-n problem where the aim is for a population of agents to identify the highest quality out of n options through local interactions between individuals and limited direct feedback from the environment. In this context we claim that possibility theory provides efficient mechanisms by which an agent can learn about the state of the world, and which can allow them to handle inconsistencies between what they and others believe by varying the level of imprecision of their own beliefs. We introduce a discrete time model of a population of agents applying possibility theory to the best-of-n problem. Simulation experiments are then used to investigate the accuracy of possibility theory in this context as well as its robustness to noise under varying amounts of direct evidence. Finally, we compare possibility theory in this context with a similar probabilistic approach.

ABSTRACT. Path planning plays a crucial role in robot action execution, since a path or a motion trajectory for a particular action has to be defined first before the action can be executed. Most of the current approaches are iterative methods where the trajectory is generated iteratively by predicting the next state based on the current state. Moreover, in case of multi-agent systems, paths are planned for each agent separately. In contrast to that, we propose a novel method by utilising fully convolutional neural network, which allows generation of complete paths, even for more than one agent, in one-shot, i.e., with a single prediction step. We demonstrate that our method is able to successfully generate optimal or close to optimal paths in more than 98\% of the cases for single path predictions. Moreover, we show that although the network has never been trained on multi-path planning it is also able to generate optimal or close to optimal paths in 85.7\% and 65.4\% of the cases when generating two and three paths, respectively.

ABSTRACT. A swarm is defined as a collection of interacting individuals or entities. It can, therefore, be represented as a network with the individuals as its nodes and the interactions as its edges. Since an individual node can only interact with its neighbourhood, it is directly aware only of the local properties of the swarm (such as its number of neighbours and its local clustering coefficient). Global properties of the network (or the swarm) such as the size of the network, its degree distribution, size and topology of the giant connected components are not directly accessible to the individual nodes of the network.

However, various collective behaviours exhibited by real-world swarms depend on the individual members being aware of the global properties of the swarm [1,2]. For example, many species of bacteria and social insects like honeybees and ants respond collectively in qualitatively different ways depending on whether the population---or population density---of their colony is above or below a certain threshold. In either case, the decisions of the colony depends on the individual members of the swarm having an estimate of the size of the swarm which is one of its global properties. Apart from swarms in biological systems, the knowledge about the size of the network is also crucial to computer and wireless networks where it determines the robustness of the network to link failures [3].

The importance of knowledge of the size of a network in determining the qualitative behaviour and robustness of the network raises an important question which we attempt to address in this study: How can all the nodes in a network determine its size in a time- and memory efficient way, while being able to communicate only with its local neighbours.

Given the importance of this question, many recent studies addressed it and proposed various algorithms to count the number of nodes in a decentralized way. Some of the proposed algorithms are stochastic in nature and rely either on (a) accumulation of local information while executing random walks [4] (b) discrete averaging of consensus on the network [3] or (c) random stimulated and non-stimulated signalling among the nodes [5]. Finally, the deterministic algorithms currently proposed for this problem rely on some prior knowledge about the structure of the network and are effective only when a single node --- and not the entire network ---wants to be aware of the network size [6].

Here we propose a deterministic algorithm for the simultaneous and decentralized computation of the network size $N$ by all its nodes with no prior assumptions about the topology of the network. The underlying idea of the algorithm is inspired by the standard node-counting method on a tree by its root. In a tree, the counts of the leaves are assimilated by their respective parents and are iteratively pruned. Applying such an algorithm on a graph with a general topology poses a challenge since a strict hierarchy does not exist among the nodes. To overcome this problem, we add a step in each iteration where, based on the degree of its neighbours, each node determines its local hierarchy which, in turn, determines whether it should be pruned or not.

The algorithm is divided into two phases: ``network reduction'' and ``count broadcasting''. The node reduction phase is an iterative process in which each node $i$ updates a local node counter $c_i$ and its state $s_i$ which can be either active ($A$) or leaf ($L$). The other two states: broadcast ($B$) and residue ($R$) are attained by the nodes in the count broadcast phase.

In the first phase, every node starts as active and initialises its local node counter $c_i=1$. In each iteration, each active node $i$ counts the number $n_i$ of its active neighbours. Then, if none of the $n_i$ neighbours has more neighbours than $n_i$, the node $i$ changes its state to leaf ($s_i = L$). A node in leaf state sends to the $n_i$ active neighbours a proportion of its counter $\frac{c_i}{n_i}$. Each node $j$ that receives a message from a node in leaf state updates its node count to $c_j = c_j + \frac{c_i}{n_i}$. At the end of each iteration, the leaf nodes change their state to broadcast ($s_i=B$) if there are active neighbours ($n_i>0$), otherwise to residue ($s_i=R$ if $n_i=0$).

Note that, while the number of active nodes in the network decreases after each iteration, the sum of $c_i$'s for all active nodes remains equal to the size of the original network. In other words, the information about the size of the network gets concentrated to a fewer number of nodes with each subsequent iteration until one or more `residue' nodes are left in the network. Each node holds partial but substantial information about the size of the network. In the count broadcast phase, the information held by each of these residue nodes ($s_i = R$) is broadcast across the network and is assimilated together by the other nodes to locally compute the complete size of the network. The complete pseudo-code for the algorithm is given in Fig. 1.

The algorithm presented here has various advantages when compared to the previously known algorithms. Since our algorithm is deterministic in nature, the results obtained by it are completely accurate in absence of node failures and computational errors. This is in contrast with the other stochastic methods where the accuracy of the results is finite for any finite time. Note that, since the all nodes do not transmit messages across the network at all times, the number of messages sent across the network is significantly lesser than the algorithms based on discrete averaging methods where each node sends messages in each iteration. The lesser number of messages transmitted results in a significant reduction in possibility of errors due to message transmission.

In our study, we analyse the robustness of the proposed algorithm to node failures and quantify its performance in terms of memory and computation time costs. In order to do so, we run the proposed algorithm on a variety of network topologies, namely, random networks, small-world networks, scale-free networks and random geometric networks. While the algorithm scales differently for different topologies, our analysis shows that it performs significantly better for networks with a more heterogeneous degree distribution due to more efficient accumulation and redistribution of information.

In this study, we also validate the proposed algorithm through its implementation on a network of Kilobots. A Kilobot is a small and low-cost robot which is designed to act in large groups. Groups of such robots are used for large-scale experiments [7]. Implementation of the algorithm in such real-world systems poses additional challenges such as lack of perfect synchrony among various nodes, limitations on the speed and amount of data transfer and the possibility of data loss. In spite of these difficulties, our initial investigations show that the current implementation of the algorithm reliably succeeds in correctly estimating the size of the network. Although currently, the speed of the algorithm is much less than the theoretical limit, it is still much faster than previously known general algorithms [5] with equivalent accuracy which have been implemented on robotic systems.

========== REFERENCES ==========

[1] K. Papenfort and B. L. Bassler, “Quorum sensing signal-response systems in Gram-negative bacteria”, Nature Reviews Microbiology, Vol. 14, pp. 576-588, 2016.

[2] H. F. McCreery and M. D. Breed, “Cooperative transport in ants: A review of proximate mechanisms”, Insectes Sociaux, Vol. 61, No. 2, pp. 99-110, 2014.

[3] S. Zhang, C. Tepedelenlioglu, A. Spanias, and M. K. Banavar, “Node counting in wireless sensor networks”, Asilomar Conference on Signals, Systems and Computers, pp. 360–364, 2015.

[4] C. Musco, H.-H. Su, and N. Lynch, “Ant-Inspired Density Estimation via Random Walks,” Proceedings of the National Academy of Sciences, Vol. 114, No. 40, pp. 10534–10541, 2017.

[5] M. Brambilla, C. Pinciroli, M. Birattari and M. Dorigo, “A reliable distributed algorithm for group size estimation with minimal communication requirements”, International Conference on Advanced Robotics, pp. 1–6, 2009.

[6] D. Dolev, O. Mokryn, and Y. Shavitt, “On multicast trees: Structure and size estimation”, IEEE/ACM Transactions on Networking, Vol. 14, No. 3, pp. 557–567, 2006.

[7] M. Rubenstein, A. Cornejo, R. Nagpal, “Programmable self-assembly in a thousand-robot swarm”, Science, Vol. 345, No. 6198, pp. 795–799, 2014.

ABSTRACT. The flock structures produced by animals or others self-organize and change their complexity over time. While it is often characterized by the spatial alignment of each element, this paper pays attention to its dynamic and hierarchical nature, its temporal variations and meta-structures. At times, the upper-structure does not change in hierarchical systems, however, its lower component changes constantly over time. Current clustering methods are intended for capturing the static and mono-layer features of complex patterning. In order to detect and track dynamic and hierarchical objects, we need a new clustering technique. Hence, in this paper, we improved the Generative topographic mapping (GTM) method to visualize such hierarchical dynamic structures as they continuously change over time. Using examples from our recent studies on the large-scale Boids model, we confirmed that this new method we developed can capture the complex flocking objects and, furthermore, that it can track the merging and collapsing events of objects.

ABSTRACT. The heap paradox is an enduring problem in complex systems. One variation of the heap paradox considers how many parts of a whole are required to generate group behaviour. The original heap paradox indicates that the recursive adding grain process finally makes a heap exceeding at a certain point; however, in this study, we insist that this recursive operation must involve ontological judgement (i.e. anticipation for the forthcoming heap from non-heap in every procedure). Applying integrated information theory (IIT) 3.0 provides two solutions to this paradox. Here, the group behaviour of P. altivelis is analysed with group sizes ranging from 2–5 fish. Resultingly, the subgroup embedded in the whole group has qualitatively different group integrity compared with a whole group of the same size: 2-fish subschools lose the followership property, but gain the leadership property, which is never observed in a 2-fish whole school. Thus, our result suggests that each fish tentatively solves the heap paradox according to its own ontological judgements concerning its group size.

ABSTRACT. Within the pelagic realm, which constitutes the largest habitat on earth, predation is especially challenging given the absence of physical hiding places. Accordingly, small marine fish turn to two anti-predator main strategies: group-living [1] and camouflage [2]. While the benefits of group-living is tightly linked to group size (e.g. increased vigilance, predator confusion, the dilution effect), camouflage is largely a property of the individual. Consequently, the two mechanisms have largely been treated separately, using different theoretical and empirical approaches. Here we show that the combination of camouflage and aggregation may, in fact, come at a cost. For most pelagic fish, the capacity to blend-in with their surrounding is achieved through specular skin-textures that mirror the ambient light [3]. However, in shallow waters, light reflected-off specular fish can result in high-intensity flashes that, in turn, may attract the attention of predators. Thus, the benefits of larger aggregates could be undermined by the increased detectability. Using models for light propagation in the ocean, our simulations couple group dynamics with the optical properties of fish and their environment (light conditions, water properties, etc.). We show that the visual imprint of the aggregate is a complex function of its size, shape and internal structure, with clear implications to detectability (see figure). Our model provides a new perspective on several behavioral attributes observed in aggregates of pelagic fish and their predators. For example, different school shapes may be optimal at different light conditions.

ABSTRACT. One of the solution for sustainable fishery resources is the fishing of well controlled fish schools. In order to realize this, it is necessary to understand the characteristics of the fish school using the analysis of water tank experiments and its numerical simulations based on the appropriate behavior model. In this study, we introduce the three-dimensional analysis and a novel behavior modelling of fish schooling. Our model is consisted of the action strategies which are generated from the environmental conditions and the state of individuals. The states have the innate characters and reflection systems based on the natural survival instinct. We develop the conditions and states for the numerical simulation and make a comparison between the water tank experimental results with the computational results. Furthermore, we try to build up the behavioral model to develop the new control method of fish schooling for the fishery.

ABSTRACT. This paper discusses on the control algorithm of a new modular robot, which is a cube-shaped robot capable of performing rotational and translational motions in multiple directions, and coupling with the other robots to form closely-packed structures. A computer simulation was performed and the result has confirmed the effectiveness of the proposed algorithm in the construction of a planar structure under micro gravity environment.

ABSTRACT. This paper deals with design of a formation control law for multiple quadrotors. The proposed approach consists of two steps. In the first step, an ideal translational control law is designed based on hierarchical linear quadratic optimal control. As a feature of this paper, the achievement of the formation is divided into an individual objective and cooperative one. The cost function comprises the corresponding performance outputs. Once the ideal control law is obtained, the attitude control law is designed so that the ideal translational motion is realized. Transient behavior can also be adjusted by changing the weights in the cost function. We exanine the effectiveness of the proposed control law by numerical simulations.

ABSTRACT. In this paper, a distributed event-triggered formation control algorithm for a multi-agent system which consists of linear discrete-time agents is proposed. For agents to achieve a formation, it is necessary to communicate with each other to feedback their information. For communication, wireless communications are desirable for various tasks since wired networks can constrain their movement. Also, each agent have to observe their states from sensors. Those energy consumption due to unnecessary calculation and communication of agents may shorten battery life of agents. By using the event-triggered protocol, which updates the control input aperiodically only when certain triggering condition is satisfied, we aim to reduce those calculation and communication by reducing the input update frequency. Also the multi-agent system is guaranteed to achieve a formation by determining the triggering condition of the control input based on Lyapunov’s stability theorem. At the end of the paper, the effectiveness of the proposed algorithm is verified by numerical simulation.

ABSTRACT. The highly synchronized parade can impress audiences how strong the troops seem to be whereas it is difficult to train for the good parade because of its complex collective behavior. However, there is no scientific research about what the important factor to train and produce for the good parade is. One of the bottleneck to the scientific approach is difficulty of measurement of a collective as same as other swarm researches.　In this paper, we measured the pose of members in the parade with OpenPose, which is a cutting-edge pose estimation technology of deep learning. By this measurement, we propose a numerical evaluation for the quality of the parade, and it is confirmed by questionnaire of themselves. In conclusion, our evaluation method is applicable for the quantitative evaluation, and it was suggested that the variation level of the arm angles was related to the quality of the parade.

ABSTRACT. Ant colonies operate without central control, using feedback from simple olfactory interactions to regulate their activities. There are more than 14K species of ants in every habitat on Earth, using diverse algorithms, producing different feedback regimes, in different environments.  These algorithms have evolved to fit the dynamics of particular environments, including operating costs,  and the threat of rupture. Ant colonies in the desert, where life is tough but stable, and in the tropical forest, where life is easy but unpredictable, use different algorithms.  Harvester ants in the desert regulate foraging effort according to current food availability and weather conditions, using the rate of contact inside the nest between returning and outgoing foragers. Turtle ants in the trees of the tropical forest use the rate of deposition of a volatile pheromone to build and maintain a distributed routing network in heterogeneous environments,  based on different design principles from the shortest paths commonly studied in ants and in network science. The diversity of ants provides opportunities to learn how collective behavior evolves to fit diverse environmental dynamics.