DARS-SWARM2021: THE 15TH INTERNATIONAL SYMPOSIUM ON DISTRIBUTED AUTONOMOUS ROBOTIC SYSTEMS 2021 AND THE 4TH INTERNATIONAL SYMPOSIUM ON SWARM BEHAVIOR AND BIO-INSPIRED ROBOTICS 2021
PROGRAM FOR WEDNESDAY, JUNE 2ND
Days:
previous day
next day
all days

View: session overviewtalk overview

04:00-06:30 Session 2A0OS5: SWARM online OS5: Multi-scale Instrumentation of Biological Swarms
04:00
Multi-Level Instrumentation of Bivouac Thermoregulation: Current Methods and Future Directions

ABSTRACT. Army ants are frequently used as charismatic organismal representatives of collective behavior in nature, providing templates for modern engineered systems as well as continuing to drive aspirational goals for the engineered systems of the future. Most attention on army ants has been focused on the ability of groups of ants to self-assemble into adaptive structures such as bridges or even whole-colony bivouacs (living nests) that provide shelter and protection. Less attention has been paid to the demonstrated abilities of these large-scale army-ant structures to thermoregulate, which is a particularly notable adaptation given that each individual ant has limited ability to regulate her own body temperature. Understanding the emergence of group-level thermoregulation from inexpensive components provides significant opportunities for engineered systems but also requires surmounting significant instrumentation challenges. Here, we review what is known about army-ant thermoregulation, what is currently possible in terms of multi-scale instrumentation, and what important gaps need to be filled to advance the field further.

04:15
Beyond Tracking: Using Deep Learning to Discover Novel Interactions in Biological Swarms

ABSTRACT. Most deep-learning frameworks for understanding biological swarms are designed to fit perceptive models of group behavior to individual-level data (e.g., spatial coordinates of identified features of individuals) that have been separately gathered from video observations. Despite considerable advances in automated tracking, these methods are still very expensive or unreliable when tracking large numbers of animals simultaneously. Moreover, this approach assumes that the human-chosen features include sufficient features to explain important patterns in collective behavior. To address these issues, we propose training deep network models to predict system-level states directly from generic graphical features from the entire view, which can be relatively inexpensive to gather in a completely automated fashion. Because the resulting predictive models are not based on human-understood predictors, we use explanatory modules (e.g., Grad-CAM) that combine information hidden in the latent variables of the deep-network model with the video data itself to communicate to a human observer which aspects of observed individual behaviors are most informative in predicting group behavior. This represents an example of augmented intelligence in behavioral ecology – knowledge co-creation in a human–AI team. As proof of concept, we utilize a 20-day video recording of a colony of over 50 Harpegnathos saltator ants to showcase that, without any individual annotations provided, a trained model can generate an "importance map" across the video frames to highlight regions of important behaviors, such as dueling (which the AI has no a priori knowledge of), that play a role in the resolution of reproductive-hierarchy re-formation. Based on the empirical results, we also discuss the potential use and current challenges to further develop the proposed framework as a tool to discover behaviors that have not yet been considered crucial to understand complex social dynamics within biological collectives.

04:30
Reconstructing Configurations of Zooplankton Aggregations and Their Biogenic Flows with 3D Scanning Velocimetry

ABSTRACT. A three-dimensional (3D) scanning velocimetry system is developed to quantify the 3D configurations of zooplankton aggregations and their surrounding volumetric, three-component velocity fields. The approach uses a translating laser sheet to rapidly scan through a volume of interest and sequentially illuminate slices of the flow containing both tracers seeded in the fluid and swimmers. These image slices are captured by a single high-speed camera, encoding information about the third spatial dimension within the image time-series. The capability of the system is characterized using a study of induced migrations of millimeter-scale brine shrimp (Artemia salina). Reconstruction of swimmer bodies and trajectories within the migrating aggregation is achieved up to the maximum number density studied presently, 8 x 10^5 animals per m^3, and is comparable to previous depth-averaged 2D measurements of similar migrations. Corresponding velocity measurements indicate that the technique can resolve the 3D velocity field in and around the swimming aggregation.

04:45
Monitoring Traffic in the Transport Networks of Arboreal Ants

ABSTRACT. In some ant species, colonies form stable networks of trails to transport food from source to nest and between nests. These transport networks effectively balance cost, efficiency and robustness—yet they emerge from the distributed action of many individuals working independently. Arboreal turtle ants are an emerging study system for the study of transport networks; they differ from well-studied ground-nesting ants in that both nesting sites and pathways between them are highly constrained by tree topology. To study the individual rules that generate arboreal networks, we must track individual movement in arenas that mimic some of the structural complexity of natural vegetation. Here, we discuss two complementary methods for tracking ant movement at targeted locations within a complex three-dimensional structure, based on microtransponders and video tracking. Together, these monitoring systems will yield insight into the way that turtle ants build networks in a constrained environment.

05:00
How to Effectively Sample to Estimate Distributions of Behavioral States and Transitions in Social Insects

ABSTRACT. Qualitative features of social insect behavior such as task performance re-quire manual tracking. Manual tracking in social insect colonies, however, can take a significant amount of time as many individuals need to be ob-served per colony. We look to optimize the way we subsample behavior by finding the optimal number of randomly chosen continuous intervals of behavior, a sampling technique we call piecewise instantaneous sampling for individuals or piecewise scan sampling for groups. We tracked the task performance of 9 harvester ants for 3 hours a piece and then simulated sam-pling from this rich dataset using differing numbers of intervals. We meas-ure two types of errors for every sample. We measure error for independent metrics of animal behavior, which are features of animal behavior which are independent of the time they are sampled (i.e. the proportion of time they spend on tasks). We also measure error for dependent metrics, which are fea-tures of animal behavior which are autocorreled in time (i.e. bout length). Including more intervals in a sample tends to minimize independent metrics whereas including fewer intervals tends to minimize dependent metrics. We find that 8 intervals optimizes the tradeoff between these types of errors in our dataset. This result seems to depend on how specialized each worker is on various tasks. As worker specialization increases, so too does the num-ber of intervals needed to estimate its behavior.

05:15
Long-Term Tracking and Classification of Individual Behavior in Bumble Bee Colonies

ABSTRACT. Social insects are ecologically dominant and provide vital ecosystem services. It is critical to understand collective responses of social insects such as bees to ecological perturbations. However, studying behavior of individual insects across entire colonies and relevant timescales (i.e., days or weeks) remains a central challenge. Here, we describe an approach for long-term monitoring of multiple entire bumble bee (Bombus spp) colonies that combines the complementary strengths of multiple methods. Specifically, we combine (a) automated, long-term monitoring, (b) tag-based individual tracking, and (c) pose estimation to quantify behavior across multiple entire colonies over a 48h period. We then develop an approach for unsupervised behavioral classification that embeds each individual’s behavior using data from both pose tracking and spatial data from within the nest. Finally, we use this approach to quantify collective behavioral response to an ecologically relevant perturbation: hours-long cold-stress.

05:30
A Customizable, Low-Cost Alternative for Distributed 2D Flow Sensing in Swarms

ABSTRACT. Many collective behaviors in social insects are mediated by airflow such as honeybees fanning their wings to drive nest ventilation or to disperse pheromones during olfactory search and aggregation. Empirical investigations of how the local sensing and actuation of individual insects scale up to produce such large scale flows require distributed flow measurement techniques. Common vision-based techniques are not workable in the cluttered dynamic environments in which these social insects live and behave. Here, we develop a customizable, low-cost 2D flow sensor that can measure both magnitude and direction and be deployed in dense sensor arrays on experimental surfaces. While many 2D thermal flow sensor designs have been published, our minimal design uses off- the-shelf components and standard fabrication techniques that should be accessible to most research groups. Here we report on the design and performance of our sensor and provide a calibration protocol that can be implemented by the user. The sensor has a measurement range of 0-2 m/s with accuracy of 0.1 m/s, angular resolution of 15◦, and a time constant of 3s. We also discuss modifications that can be made to tune sensor performance for a given application.

05:45
Use of a Customizable, Disassemblable X-Ray CT Scanner for in-the-Field Imaging of Social Insect-Made Structures

ABSTRACT. Social insects form complex three-dimensional structures, both from the environment (in the case of excavated or constructed nests) and from their own bodies (in the case of self-assemblies). Because these structures or the animals constructing them can often not be transported or maintained in a laboratory setting, methods to study such structures in the field are required. Here we introduce a customizable, disassemblable X-ray Computed Tomography (CT) scanner that can be transported to field sites and can func-tion in challenging environmental conditions. We employ a novel marker-based live-calibration method that makes the scanner setup robust to physical inaccuracies resulting from transport or in-the-field handling and operation. We discuss use-cases for the study of social insects in the field and illustrate them with examples of three-dimensional reconstructions of individual animals and animal-built structures.

06:00
Automated Entrance Monitoring of Managed Bumble Bees

ABSTRACT. Social pollinators are a critical part of our ecosystem and a great source of inspiration for engineered swarms. Recently, researchers have produced a range of systems for automated monitoring of honey bee entrances to further insights on e.g. collective foraging, labor distribution, and suppression of disease transmission. In this article, we detail the design of a system customized for capturing top and side view photos of bumble bees as they enter and exit their hives. We show how these photos can be used to automatically track foraging activity, identify individuals, and characterize bee size and pollen presence. To aid technology adoption by biologists, our design is specifically optimized for low cost, easy fabrication, operation, and maintenance. Over two iterations, the entrance has been used on 6 hives in greenhouse and field over 6 weeks.

06:15
Fire Ant Rafts Are Unstable

ABSTRACT. Fire ants survive flash floods by linking their bodies together to build waterproof rafts. How ants find each other on the water surface and bind together is still poorly understood. In this experimental study, we measure the interactions between ants on the water surface. Raft building is accomplished through a competition between the ants' tendency to explore by taking random walks and the Cheerios effect, which brings them together by capillary forces. We present statistics for their mean square displacement, showing that their motion is neither diffusive nor ballistic. Even when ants come into contact they repel each other with some probability. The combined effects of capillarity and random walks lead to a minimum size of 100 ants for a raft to be stable. The ants' decentralized methods of raft formation may provide insights into the design of other self-assembling robotic systems.

10:00-12:00 Session 2A1S: SWARM online 1
10:00
Dynamics of Cooperative and Non-Cooperative Relationships Built Through Indirect Reciprocity in a Realistic Acquaintance Network

ABSTRACT. Opinion dynamics based on an indirect reciprocity is studied on a realistic network structure. We conduct so-called donation games in a human communication network which we obtained from mobile communication logs. We find two communities which have significantly high opinion-change frequency in this network. This study suggests that the opinion-change frequency of a couple depends on the community structure the couple belongs to.

10:30
On the Distance Distribution from Captured Targets to the Chasers in Group Chase and Escape

ABSTRACT. It has been suggested that the presence of lazy individuals may improve the efficiency of the entire herd of chasers in the group chase and escape. However, it remains unclear how to maintain moderate laziness in the group. In this study, we investigate the distance distribution from captured targets to the diligent and lazy chasers. The results show that the lazy chasers are close to the captured targets in the late phase of group chase and escape. It suggests that there is a possible benefit of being lazy individuals.

11:00
OS3: Consensus Control of Multi-Hopping-Rover Systems

ABSTRACT. Hopping rover (called Hedgehog) is a space exploration robot that can move in low-gravity environments such as small solar system bodies. Hopping rovers can freely travel in a rough environment without getting stuck in obstacles. By taking this advantage, they are expected to perform various missions in unknown and harsh environments. Moreover, by organizing a group of hopping rovers, their applicability will be enhanced. In this paper, we address multi-hopping-rover systems and present a framework of consensus control. In particular, we derive a distributed controller, which establishes the supermartingale property for a performance index quantifying the degree of consensus.

11:30
OS3: Design of an Experimental Platform to Control Simple Swarm Robots from Environmental Stimuli

ABSTRACT. The purpose of this paper is to improve the performance of a swarm robot system by designing stimuli from the environment while the robots remain simple and cheap. As a first step, this paper designs the 3 input patterns using the light stimuli which affects the robot movements. Specifically, we experimentally verified that by designing light stimuli from the environment, we can realize a formation in which the robots moves along a course at constant speed.

10:30-12:00 Session 2A1D: DARS online 3
10:30
ReactiveBuild: Environment-Adaptive Self-Assembly of Amorphous Structures

ABSTRACT. ReactiveBuild is an algorithm that enables swarms of robots to build a variety of robust, environment-adaptive structures without pre-planning. Robots form structures by climbing their peers until either reaching a point closest to a goal location or until a neighboring robot recruits it for structural reinforcement. This contrasts with typical approaches to robotic self-assembly which generally seek to form some a priori shape. This paper demonstrates a simulated swarm of Fire-Ant3D robots using ReactiveBuild to form towers, chains, cantilevers, and bridges in three-dimensional environments.

11:00
Collective Transport via Sequential Caging

ABSTRACT. We propose a decentralized algorithm to collaboratively transport arbitrarily shaped objects using a swarm of robots. Our approach starts with a task allocation phase that sequentially distributes locations around the object to be transported starting from a seed robot that makes first contact with the object. Our approach does not require previous knowledge of the shape of the object to ensure caging. To push the object to a goal location, we estimate the robots required to apply force on the object based on the angular difference between the target and the object. During transport, the robots follow a sequence of intermediate goal locations specifying the required pose of the object at that location. We evaluate our approach in a physics-based simulator with up to 100 robots, using three generic paths. Experiments using a group of KheperaIV robots demonstrate the effectiveness of our approach in a real setting.

11:30
Processes for a Colony Solving the Best-of-N Problem Using a Bipartite Graph Representation

ABSTRACT. Agent-based simulations and differential equation models have been used to analyze distributed solutions to the best-of-N problem. This. paper shows that the best-of-N problem can be also solved using a graph-based formalism that abstractly represents (a) agents and solutions as vertices, (b) individual agent states as graph edges, and (c) agent state dynamics as edge creation (attachment) or deletion (detachment) between agent and solution. The paper identifies multiple candidate attachment and detachment processes from the literature, and then presents a comparative study of how well various processes perform on the best-of-N problem. Results not only identify promising attachment and detachment processes but also identify model parameters that provide probable convergence to the best solution. Finally, processes are identified that maybe suitable for the best-M-of-N problem.

12:40-14:40 Session 2P1D: DARS online 4
12:40
Decentralized Navigation in 3D Space of a Robotic Swarm with Heterogeneous Abilities

ABSTRACT. This paper proposes a decentralized method for navigation of multiple robots, each of which has different abilities, by a single leader robot in 3D space, especially focusing on the connectivity maintenance. We assume a swarm of robots whose sensing ranges, maximum speeds and maximum accelerations are different. For such robots, we propose a control method for maintaining the whole connectivity by each agent’s keeping local connectivity in a decentralized way. We also mathematically prove that the proposed method can enable multiple robots to navigate in 3D space while keeping the connectivity. Finally, numerical simulation results are presented to confirm the effectiveness of the proposed method.

13:10
Using Reinforcement Learning to Herd a Robotic Swarm to a Target Distribution

ABSTRACT. In this paper, we present a reinforcement learning approach to designing a control policy for a ``leader'' agent that herds a swarm of ``follower'' agents, via repulsive interactions, as quickly as possible to a target probability distribution over a strongly connected graph. The leader control policy is a function of the swarm distribution, which evolves over time according to a mean-field model in the form of an ordinary difference equation. The dependence of the policy on agent populations at each graph vertex, rather than on individual agent activity, simplifies the observations required by the leader and enables the control strategy to scale with the number of agents. Two Temporal-Difference learning algorithms, SARSA and Q-Learning, are used to generate the leader control policy based on the follower agent distribution and the leader's location on the graph. A simulation environment corresponding to a grid graph with 4 vertices was used to train and validate the control policies for follower agent populations ranging from 10 to 1000. Finally, the control policies trained on 100 simulated agents were used to successfully redistribute a physical swarm of 10 small robots to a target distribution among 4 spatial regions.

13:40
Preservation of Giant Component Size After Robot Failure for Robustness of Multi-Robot Network

ABSTRACT. This paper approaches a network topology control method for networked multi-robot systems. Although robustness of network connectivity against robot failures is a matter of concern for the multi-robot control, the robustification impedes the motion of robots because of limitations of the wireless communication. For mitigating the impediments, we focus our attention on the giant connected component size after a single robot fails, and aim to control such component size. A modified algebraic connectivity is introduced here as an indicator of the component size: a threshold for the algebraic connectivity is analyzed to preserve the component size. Theoretical properties and numerical examples are shown to demonstrate our control method.

14:10
Swarm Localization Through Cooperative Landmark Identification

ABSTRACT. In this paper we propose a landmark-based map localization system for robotic swarms. The proposed system leverages the capabilities of a distributed landmark identification algorithm developed for robotic swarms presented in [1]. The output of the landmark identification consists of a vector of probabilities that each individual robot is looking at a particular landmark in the environment. In this work, this vector is used individually by each component of the swarm to feed the measurement update of a particle filter to estimate the robot location. The system was tested in simulation to validate its performance.

12:40-14:40 Session 2P1S: SWARM online 2
12:40
OS3: Proposal of Minimal Approach on Swarm Navigation with Heterogeneity

ABSTRACT. The propose of this paper is to construct a new navigation systemfor swarm robots that does not require complicated calculations. In this paper,we propose the navigation scheme in which group robots are navigated only bycollisions from a small number of heterogeneous robots. Furthermore, we assumethat the navigator robot guides base on a planer-view : the robot cannot identifyeach worker robot but recognizes the group as a mass. We construct the model ofeach robot and conduct numerical simulations to verify the proposed controllers.

13:10
OS3: Shepherding Algorithm for Agents with Heterogeneous Sensitivities

ABSTRACT. In this abstract, we briefly report our recent work for shepherding heterogeneous flocks. We consider the heterogeneity in the sensitivity of the sheep against the sheepdog. The proposed shepherding algorithm iteratively uses an existing algorithm called Farthest Agent Targeting. We report by numerical simulations that the proposed algorithm can outperform Farthest Agent Targeting.

13:40
Analyzing Predation Risk Through Spatial Locations in Wild Anubis Baboons (Papio anubis)

ABSTRACT. Predation is considered to be a major factor in the evolution of group living. However, predation risk has been rarely studied in groups of animals, alt-hough several measures have been proposed over the years. This is because of difficult observation during daytime travelling and because human ob-servers influence predator-prey dynamics. By using GPS collars, we studied the predation risk of wild anubis baboons (Papio anubis) based on their spatial location: unlimited and limited domains of danger, circular variance, positions based on the minimum convex polygon, and distances among in-dividuals. There was an agreement between all metrics, suggesting that male adults were more isolated and peripheral, therefore presumably, were ex-posed to higher predation risk. No difference in any predation risk metric was observed between all other sex and age class. We briefly discuss the advantages and disadvantages of each predation risk metric in relation to the spatial distribution under study.

14:10
Localization of a Snake-like Robot with the Position Information of Links in Contact with the Ground

ABSTRACT. Snake-like robots are expected to be used as rescue robots because of their ability of moving in narrow spaces and on uneven terrain. In such environments, an operator has a difficulty in recognizing the position of the robot in its surroundings. In order to solve this problem, we propose a method of localization of the robot. The estimated self-position information can be used for showing the amount of robot movement to the operator. The proposed method relies on the history of joint angle information of the snake-like robot. As a result of evaluating the proposed method using the physics engine, we can estimate the traveling length of the robot with a small error.

14:50-16:50 Session 2P2S: SWARM online 3
14:50
Analysis of Swarm Deformability Change with Control Barrier Function

ABSTRACT. In this paper, we analyzed responses against perturbations of swarming robots that were equipped with Control Barrier Function (CBF). By introducing CBF, we can guarantee safety distance for mobile robots. However, these inequality constraints can affect deformation of swarm. By applying normalized force and measuring response of robots, we investigate effects for deforming by CBF constraint. Using this result, we can get an appropriate range of parameter which realizes both keeping distance and retaining deformability.

15:20
Cooperative Movement Based Path Planning with a Mobile Robot in a Multi-Agent Environment Using Deep Reinforcement Learning

ABSTRACT. This study proposes a simulation framework on cooperative movement between a mobile robot and a small crowd of pedestrians using deep reinforcement learning. To achieve this goal, the framework will utilize mutual concession as well as other forms of inducement such as voice reminders and gentle physical contact. The scenario features a path navigation setting whereby a mobile robot is tasked to maneuver towards a predetermined destination and back while also avoiding pedestrians along the way. Meanwhile, the robot is observed to determine the effectiveness of each of the three inducement methods based on three primary attributes. The three evaluated attributes being the effective distance, politeness, and the overall success rate. Subsequently, the robot and pedestrian models are designed with discrete movement and action features to interact within the framework. Finally, we developed a model based on Deep Q-Network and trained the robot in the scenario with the state and reward system. Additionally, we designed two baselines to verify the importance of the pedestrian model characteristics and the effectiveness of the inducement methods. In the 1st experiment, the results show that the pedestrian’s active avoidance probability and the pedestrian’s field of view have a large influence on the robot’s actions. The results of the 2nd experiment convey that the introduction of voice reminder s and physical contact effectively reduces the collision count and arrival time.

15:50
Using Deep Learning for Rehabilitation Exercise Based on Brain-Machine Interaction with Wearable Robot

ABSTRACT. Objective: This study proposes deep learning called convolutional neural network (CNN), which can enhance the classification performance of five different rehabilitation exercise scenarios based on brain-machine interaction (BMI). Also , it further presents the feasibility of helping stroke patients and healthy people to assist in rehabilitation exercise. Method: This work designs a wearable hand robot and five different motor images (MI) for exercise guidance on the computer screen. A participant is also asked to set on a chair to acquire the cerebral response signals using the function near-infrared spectroscopy (fNIRS). Deep learning called convolutional neural network (CNN) is utilized to extract and classify the collected data and make commands to the wearable hand robot. Results: The classification accuracy of the S_2 MI is the highest value for participant 1; the classification accuracy of the S_1 MI is the highest value for participant 2. Besides, the S_5 and S_3 MI showed the lowest classification accuracy for participant 1 and participant 2, respectively. Significant: Propose a CNN framework with visual guidance to control wearable robots to reduce incorrect commands, and present a rehabilitation exercise feasibility to stroke patients and healthy people with less time.

16:20
Concept of a Description-Based Approach to Represent Manipulation Actions in Multi-Agent Interaction

ABSTRACT. Swarm robotics is an approach to the coordination of multiple robots as a system which consist of several (often simple) physical robots. This approach assumes that a desired collective behavior emerges from the interactions between a robot with the other robots and with the environment. Agents in a swarm can show different levels of behavioral complexity. Here we are considering agent that can purposefully act. These could be ants (ant robots), that can manipulate objects but also other artificial agents that can do this up to the level of humanoids. In all these cases cooperation is only possible when the machines are able to represent and recognize interactions. This is especially important when interacting also with humans, because in this case direct signal transmission between agents is not possible. In this paper, we focus on multi-agent interactions in manipulation action applications and present a framework for representing inter-agent interactions.

17:00-18:00 Session 2SP1: Plenary Speech 1
17:00
Controlling Cell Fate Specification System Based on Network Structure

ABSTRACT. By the success of modern biology we have many examples of large networks which describe regulatory interactions between a large number of genes. On the other hand, we have a limited understanding for the dynamics of molecular activity based on such complex networks. To overcome the problem, we developed Linkage Logic theory by which important aspects of dynamical properties are determined from information of the regulatory linkages alone. The theory assures that i) any long-term dynamical behavior of the whole system can be identified/controlled by a subset of molecules in the network, and that ii) the subset is determined from the regulatory linkage alone as a feedback vertex set (FVS) of the network. We applied the theory to the gene regulatory network for cell differentiation of ascidian embryo, which includes more than 90 genes. From the analysis, dynamical attractors possibly generated by the network should be identified/controlled by only 5 genes, if the information of the network structure is correct. We verified our prediction by combinatorial experiments of knockdown and overexpression by using ascidian embryos. We found that almost all of the expected cell types, six out of seven major tissues, could be induced by experimental manipulations of these 5 genes.

18:10-19:10 Session 2SP2: Invited Speech 1
18:10
Modeling and Control of Underwater Snake Robots

ABSTRACT. In this talk, I will present how inspiration from nature has led to the development of snake robots. I will present the mathematical models of such robots and look into how certain properties of the model corresponds to the properties of snake movement. The connection between certain parameters of the gait and the forward velocity of the robot will also be discussed. Different control methods for snake robot path following, and the corresponding stability analysis will then be discussed. I will show several of our snake robots, ranging from land based university prototypes to commercial underwater swimming manipulators.