In recent years, machine-learning applications have been rapidly expanding in the fields of robotics and swarm systems, including multi-agent systems. Swarm systems were developed in the field of robotics as a kind of distributed autonomous robotic systems, imbibing the concepts of the emergent methodology for extremely redundant systems. They typically consist of homogeneous autonomous robots, which resemble living animals that build swarms. Machine-learning techniques such as deep learning have played a remarkable role in controlling robotic behaviors in the real world or multi-agents in the simulation environment.
In this special issue, we highlight five interesting papers that cover topics ranging from the analysis of the relationship between the congestion among autonomous robots and the task performances, to the decision making process among multiple autonomous agents.
We thank the authors and reviewers of the papers and hope that this special issue encourages readers to explore recent topics and future studies in machine-learning applications for robotics and swarm systems.
Swarm robotic systems (SRSs) are a type of multi-robot system in which robots operate without any form of centralized control. The typical design methodology for SRSs comprises a behavior-based approach, where the desired collective behavior is obtained manually by designing the behavior of individual robots in advance. In contrast, in an automatic design approach, a certain general methodology is adopted. This paper presents a deep reinforcement learning approach for collective behavior acquisition of SRSs. The swarm robots are expected to collect information in parallel and share their experience for accelerating their learning. We conducted real swarm robot experiments and evaluated the learning performance of the swarm in a scenario where the robots consecutively traveled between two landmarks.
This paper focuses on the effect of congestion on swarm performance by considering the number of robots and their size. Swarm robotics is the study of a large group of autonomous robots from which collective behavior emerges without reliance on any centralized control. Due to the fact that robotic swarms are composed of a large number of robots, it is important to consider the congestion among them. However, only a few studies have focused on the relationship between the congestion and the performance of robotic swarms; moreover, these studies only discuss the effect of the number of robots. In this study, experiments were conducted by computer simulation and carried out by varying both the number of robots and the size of the robots in a path formation task. The robot controller was designed with an evolutionary robotics approach. The results show that not only the number of robots but also their size are essential features in the relationship between congestion and swarm performance. In addition, autonomous specialization within the robotic swarm emerged in situations with moderate congestion.
In this paper, the mission for mobile patrolling robots is to detect as many incoming visitors as possible by monitoring the environment. For multi-robot mobile patrolling systems, task assignment in the common environment is one of the problems. For this problem, we use a territorial approach and partition the environment into territories. Thus, each robot is allowed to patrol a separate territory regardless of the others. In this regard, however, the workload balancing of the patrolling tasks in the territories is a challenge. For this challenge, we propose dynamic partitioning strategies focusing on visitor trends. The system transfers a part of the territory with the maximum workload to others so as to equalize the workloads. As a result, while the sizes of the territories without visitor trends increase, others with the trends decrease. Therefore, the territorial approach enables robots to intensively monitor areas in accordance with the number of the visitors. This is the main contribution of this paper. Simulation experiments show that the patrolling robots successfully detect visitors through workload balancing.
This study deals with the estimation of the changes that occur in the Business-to-Business (B2B) networks in the Japanese textile and apparel industry by applying datasets of about 2000 companies from 2011/2012 to 2015/2016. Network analysis was used to examine the properties of the B2B networks. A factor of innovation in information and communications technology (ICT) and logistics technology was introduced into an agent-based model to demonstrate changes occurring in the related structures of B2B networks. The agent-based model was designed and tested based on qualitative information on Japanese textile and apparel industries. Consequently, network analysis revealed power-law properties and the structures of centralized hub companies. Moreover, in the simulation experiments, the centralizations of the networks generated by the agent-based model due to innovation in ICT and logistics technology were illustrated. Therefore, one of the predicted cases regarding changes that occur in the B2B networks was explained as centralizations to hub companies.
The best-of-n problem (BSTn) is a collective decision-making problem in which a swarm of robots needs to make a collective decision about a set of n choices; specifically, to decide what choice offers the best alternative . The BSTn captures the structure and logic of the discrete consensus achievement problems that appear in several swarm robotics scenarios. Although numerous algorithms have been proposed recently to deal with more than two choices, the number of choices that can be dealt with is not large. The bias and raising threshold (BRT) algorithm proposed by Phung et al.  enables swarms to deal with a large number of choices (n≫2). However, the algorithm’s goodness has not been evaluated in any practical problems, and it is necessary to evaluate the algorithm in a problem where a large number of choices exist. In this paper, we consider the best of proportions (BOP) problem; that is a version of BSTn in which a large number of choices can be dealt with by adjusting the values of different proportions. In previous research on swarms that needed to solve the BOP problem, there is only a study on the response threshold models for the division of labor. In the present study, we investigate a scenario of the BOP and apply the BRT algorithm to find the best proportion. In our previous work , a fixed proportion setting method has been adopted. Here, we adopt a stochastic proportion setting method to verify the relationship between the efficiency and the number of choices in a more general case. The results show that with a larger number of choices, the decision making becomes more efficient with high equality; that is a result that has not been found in .
The trochoid is a geometrically complete solution for realizing omnidirectional mobility with a rotating mechanism. The proposed mechanism is implemented as a novel omnidirectional vehicle with following a geometrically complete trochoidal trajectory. Because the mechanism has a large camber angle, it has an improved ability to travel past rough terrain as compared to a regular vehicle with regular wheels. In this paper, a complete mechanical control using link mechanism to generate not only the steering angle and camber angle for an ideal trochoidal wave but also the angular velocity of the wheel axis is proposed.
To achieve the control of a small-sized robot manipulator, we focus on an actuator using a shape memory alloy (SMA). By providing an adjusted voltage, an SMA wire can itself generate heat, contract, and control its length. However, a strong hysteresis is generally known to be present in a given heat and deformation volume. Most of the control methods developed thus far have applied detailed modeling and model-based control. However, there are many cases in which it is difficult to determine the parameter settings required for modeling. By contrast, iterative learning control is a method that does not require detailed information on the dynamics and realizes the desired motion through iterative trials. Despite pioneering studies on the iterative learning control of SMA, convergence has yet to be proven in detail. This paper therefore describes a stability analysis of an iterative learning control to mathematically prove convergence at the desired length. This paper also details an experimental verification of the effect of convergence depending on the variation in gain.
Development of human resources in the field of science and technology is a necessity in the modern world. To encourage the development of the nation, it is important to educate forward-thinking engineers in cooperation with the education system, with the requirements of modern society as the background of their efforts. In Western countries, STEM (Science, Technology, Engineering, and Mathematics) education is provided as a national strategy for development, and this type of education is also spreading to Japan. On the other hand, in the context of high school/university articulation, it is argued that the abilities required by high schools, universities, and society are different from one another, and that it is necessary to inter-link them. Accordingly, comprehensive content that includes all levels, from primary to higher education, is required. On the basis of this background and the current problem of STEM education and high school/university articulation, this research aims to aid in the development of a curriculum and educational study materials through which the technology of universities could be integrated into high school classes. This report includes the development of educational study material on biological information systems because it reflects the features of the department to which the authors belong and there is already a lot of research on, and technology related to, biological information. Using the educational study material, we carry out an in-class practice session including task-oriented research at a high school with cooperation from a university and discuss the educational effect that results from it.
Structure from Motion (SfM), as a three-dimensional (3D) reconstruction technique, can estimate the structure of an object by using a single moving camera. Cameras deployed in underwater environments are generally confined to waterproof housings. Thus, the light rays entering the camera are refracted twice; once at the interface between the water and the camera housing, and again at the interface between the camera housing and air. Images captured from scenes in underwater environments are prone to, and deteriorate, from distortion caused by this refraction. Severe distortions in geometric reconstruction would be caused if the refractive distortion is not properly addressed. Here, we propose a SfM approach to deal with the refraction in a camera system including a refractive surface. The impact of light refraction is precisely modeled in the refractive model. Based on the model, a new calibration and camera pose estimation method is proposed. This proposed method assists in accurate 3D reconstruction using the refractive camera system. Experiments, including simulations and real images, show that the proposed method can achieve accurate reconstruction, and effectively reduce the refractive distortion compared to conventional SfM.
This paper proposes a new control method for musculoskeletal systems, which combines a feed-forward input with a feedback input, while considering an output limit. Our previous research proposed a set-point control that used a complementary combination of feedback using a time delay and a muscular internal force feed-forward; it achieved robust and rapid positioning with relatively low muscular contraction forces. However, in that control method, the range of motion of the musculoskeletal system was limited within a horizontal plane. In other words, that system did not consider the effect of gravity. The controller proposed in this paper can achieve the reaching movement of the musculoskeletal system without requiring accurate physical parameters under gravity. Moreover, the input of the proposed method can be prevented from becoming saturated with the output limit. This paper describes the design of the proposed controller and demonstrates the effectiveness of the proposed method based on the results of numerical simulations.
Towards improving the stability of point-foot biped robot on slippery downhill, a novel and indirect control method is introduced in this paper using active wobbling masses attached to both legs. The whole dynamics which contains walking, sliding and wobbling, can be dominated by high-frequency oscillation via entrainment effect. Stable gaits are therefore achieved by controlling only 1% of the whole system where the original passive dynamic walking fails. First, we derive the equations of dynamics and control for this indirectly controlled biped walking on slippery downhill. Second, we numerically show the possibility of improving the stability with high-frequency oscillation. We also find the main effect of wobbling motion on walking via phase-plane plot. Third, we prove that the range of stable walking with respect to frictional coefficient can be enlarged by employing suitable high-frequency oscillation via parametric study. Our method will be further applied to more general conditions in real tasks which contain different locomotion types, where the whole dynamics could be dominated by high-frequency oscillation and the phase properties of the dynamics will be positively utilized.
In the normal course of human interaction people typically exchange more than spoken words. Emotion is conveyed at the same time in the form of nonverbal messages. In this paper, we present a new perceptual model of mood detection designed to enhance a robot’s social skill. This model assumes 1) there are only two hidden states (positive or negative mood), and 2) these states can be recognized by certain facial and bodily expressions. A Viterbi algorithm has been adopted to predict the hidden state from the visible physical manifestation. We verified the model by comparing estimated results with those produced by human observers. The comparison shows that our model performs as well as human observers, so the model could be used to enhance a robot’s social skill, thus endowing it with the flexibility to interact in a more human-oriented way.