Recently, mobility technologies have undergone remarkable advances, with innovations such as autonomous driving and vehicle electrification, becoming increasingly integrated into society. The deployment of vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communications has accelerated the realization of connected cars, and the use of advanced driver assistance systems (ADAS) powered by artificial intelligence (AI) and machine learning continues to expand. Simultaneously, the emergence of mobility as a service (MaaS) and the introduction of smart infrastructure are becoming essential to ensure safe and efficient urban transportation. Collectively, these developments have reshaped the concept of mobility and driven the global transition toward autonomous driving.

This special issue on “Vehicle and Mobile Robot Technology” consists of two issues (Vol.37 No.5 and No.6). The preceding issue (Vol.37 No.5) presented 20 selected papers organized into four technical categories: reinforcement learning and machine learning; control theory and algorithms; sensing, simulation, and system identification; and human–machine interfaces.

The present issue (Vol.37 No.6) further extends this exploration by presenting 19 papers that reflect the growing diversity and sophistication of research in vehicle and mobile robot systems. The contributions in this issue are organized into five thematic categories:

• Navigation and path planning

• Localization, SLAM, and environmental perception

• Control systems

• Cooperative control and multi-robot systems

• Human behavior modeling and traffic dynamics

These categories capture the central areas of emphasis in contemporary mobility research, encompassing advances in autonomous navigation and scene understanding, improvements in sensing and localization under real-world conditions, progress in vehicle and robot control methodologies, the growth of multi-agent and swarm-based approaches, and the integration of human behavior models into transportation and mobility systems. Together, these works illustrate the ongoing evolution of technologies that support safe, intelligent, and adaptive mobility. We would like to express our sincere appreciation to all the authors for their valuable contributions, as well as the reviewers for their diligent efforts in ensuring the quality of this issue. We hope that this special issue will stimulate further research and development and offer new insights into the future of vehicle and mobile robot technology.

The mainstream approach employing light detection and ranging (LiDAR) estimates the self-position of mobile robot by matching the point cloud acquired during navigation with that recorded in advance, in order to autonomously navigate to the goal point. However, this method is problematic in that it is vulnerable to environmental changes and that much effort and expenses are required to construct and update the point cloud map. Thus, in this paper, we propose an autonomous navigation method that does not require constructing a point cloud map by visiting the site in advance and is robust against environmental changes. The proposed method carries out autonomous navigation by using RTK-GNSS, and deep-learning algorithm of semantic segmentation and YOLO, A^* algorithm for path planning, and pure pursuit algorithm for path navigation. Furthermore, obstacle avoidance is carried out using semantic segmentation, YOLO, and kernel density estimation. We conducted a navigation experiment, in which a 300 m section was autonomously navigated, thus verifying the validity of proposed method.

Most of the developed mobile robots are equipped with LiDAR sensors and use SLAM for localization to move autonomously. Moreover, when operating a mobile robot using SLAM, changing the route requires specialized knowledge, and not everyone can operate it. Mobile robots are expected to patrol a predetermined route within a closed indoor space, such as security guards or transporting goods. In such cases, autonomous traveling is possible if the robot can recognize the route it is traveling without creating a map of the environment, as with SLAM. In this study, we developed an autonomous control system that uses camera images mounted on a mobile robot to calculate a straight path based on the distinct recognition of floor and wall surfaces using instance segmentation with deep learning. According to the result that the recognition area of the floor surface expanded in the direction of the branching in the branched path, it was determined that the branching was possible. In addition, when traveling along a path with a wall in front of it, a problem occurred because the target path could not be generated owing to the loss of the floor surface in the recognition results. Therefore, we recognized the wall surface using point-cloud processing and generated a target path. The proposed system allows a mobile robot to autonomously patrol a simple route, such as a corridor, by simply specifying the patrol path, such as straight ahead or turning left or right. Autonomous running experiments were conducted on a mobile robot in a corridor, including a junction point, to verify the effectiveness of the proposed system. This method allows autonomous route patrolling by a mobile robot indoors, without requiring specialized knowledge, such as SLAM, and can also be used to change routes.

This paper proposes a method for detecting and recovering from immobilized states in mobile robots operating in outdoor environments. In recent years, the demand for autonomous mobile robots capable of navigating outdoor environments has been increasing. In particular, the delivery industry anticipates replacing human-driven transport with robots to cover the section between pickup and delivery points. However, achieving autonomous navigation in outdoor environments is significantly more challenging than in indoor settings as robots must coexist with pedestrians and vehicles while complying with traffic regulations, such as obeying pedestrian signals. In this study, we aimed to develop a mobile robot capable of autonomous navigation from a starting point to a target point in a real outdoor environment. We implemented a self-position estimation method using scan matching between the real-time point-cloud data and a loop-closure-corrected three-dimensional point-cloud map. This was integrated with a path-planning system that combines detection and recovery of a robot from immobilized states, prioritized velocity control, and event-based waypoints, enabling safe autonomous navigation that complies with traffic rules in outdoor environments. In addition, we implemented a control architecture that prioritizes asynchronous velocity commands sent in parallel. To enhance the autonomy of mobile robots, a method was proposed in this study in which the robot detects obstacles and recovers from an immobilized state when it becomes stuck after colliding with these obstacles that cannot be captured by obstacle sensors during navigation. To evaluate the effectiveness of the proposed method, we conducted experiments using an autonomous mobile robot developed in this study to recover it from being stuck on a low step that the obstacle sensor could not detect. In the final run in the Tsukuba Challenge 2024, the system successfully performed four tasks of signal recognition and crossings and completed visits to both pickup and delivery destinations. The mobile robot also completed a 2 km course and was awarded the Tsukuba Mayor’s Award.

An online-usable digital twin system integrated into a control system is proposed and developed in this study to control an autonomous mobile robot operating on a complex three-dimensional (3D) terrain that includes slopes and uneven surfaces. The system is equipped with validated functions that are necessary for virtual or physical autonomous navigation. In this development, (1) 3D virtual environment, 3D virtual robot model, and 3D virtual sensor models are constructed to accurately reproduce the motion of a mobile robot on a computer. Additionally, (2) by connecting these virtual and physical models through an interface, we integrate them as a digital twin and implement an autonomous navigation control system that is synchronized with a motion simulation by switching between a physical and virtual robot (through a physics engine). This allows the same control system to be applied to both the virtual and physical models. The virtual model is created using the Unity 3D game engine, which integrates environmental terrain data and robot models to enable 3D physical simulations including road slopes and unevenness. Additionally, a path-setting method using Bézier curves and a path-following algorithm are implemented in the simulation system. Autonomous navigation of the mobile robot through the digital twin system is achieved by combining the functions above in a manner that allows online use during operation. Finally, autonomous navigation tests are conducted in a physical environment to confirm the effectiveness of the developed system.

With the rapid development of three-dimensional sensors, such as LiDAR, there is an increasing demand for accurate motion estimation from point cloud data in dynamic tasks like autonomous driving and robot navigation. To address the limitations of traditional methods in terms of efficiency and accuracy when handling sparse point clouds containing multiple objects, non-rigid motion, and noise, this paper presents an unsupervised spatio-temporal gradient flow estimation framework, called Spatio-Temporal Gradient Flow (STG-Flow). Unlike traditional methods, this approach does not rely on large labeled datasets or assume rigid-body motion. STG-Flow segments continuous-frame point clouds by combining global density statistics with supervoxel clustering. It then adaptively adjusts clustering parameters using an upper and lower bound filtering mechanism to mitigate the effects of extreme cases. After segmentation, optical flow refinement is applied to each local cluster using spatio-temporal gradient constraints, along with a multi-level robust optimization strategy and domain grouping. This method enhances the stability and accuracy of motion estimation, even under large displacements. Experiments demonstrate that STG-Flow achieves more accurate motion predictions for local object-level motion estimation in sparse scenarios. Its registration accuracy is comparable to the iterative closest point method, while offering approximately ten times higher computational efficiency, showcasing strong real-time performance and robustness.

As a safety measure against the communication risk caused by the use of the cloud for self-position estimation by autonomous mobile robots, we propose a method that combines the accurate 3D self-position estimation FAST-LIO on the cloud side and the minimum 2D self-position estimation AMCL on the robot side. In this study, we created an environment in which communication delays and disruptions are added by software. We then examined how these affect the self-location estimation of FAST-LIO in the cloud. In addition, to improve the reliability of the overall system, the advantages of each algorithm were leveraged and their disadvantages complemented by effectively combining the results of different self-position estimations (FAST-LIO in the cloud and AMCL on the robot side) using the unscented Kalman filter (UKF). The experimental results showed that stable self-position estimation at 100 ms intervals can be achieved using the UKF to combine AMCL (which is updated at 100 ms intervals on the robot side) with FAST-LIO on the cloud side (where update times are at worst 1 s due to latency and other factors).

A filtering method for observable point clouds from travel path is proposed to improve the efficiency of each point in the point cloud map for self-localization of autonomous mobile robots. The method involves retaining only the point clouds along the planned travel path of the mobile robot that can be observed by it. These points constitute the observable point cloud map. By performing ray casting from the position of the depth sensor along the travel path, observable regions are extracted whereas unobserved point clouds, such as those behind obstacles or out of the sensor range, are removed. The effectiveness of the method is evaluated through comparative experiments involving self-localization using both an original wide-area point cloud map and the observable point cloud maps. A new metric called localization contribution per point is introduced to quantify the contribution of each point, in the point cloud map, to self-localization. The experimental results demonstrate the efficiency of the observable point cloud map when used for self-localization.

Global and local localizations are critical for mobile robots to perform tasks without human support. In this study, we proposed a new method for global and local localizations that combines an environmental magnetic field with a scan matching method. In this method, we record the environmental magnetic field at each waypoint during the map-building process. To perform global localization, we rotate the mobile robot’s magnetic sensor-detected heading direction (current magnetic heading direction) to approximate the heading direction of the environmental magnetic field stored on the map. Subsequently, we use the scan matching method at positions around each waypoint. The mobile robot is currently in the position with the highest matching rate. The experimental results show that the mobile robot can successfully localize almost anywhere in the indoor environment, with the exception of a position where the magnetic sensor is affected by electric boards, and has a high success rate in an outdoor environment. For the local localization, we calculate the deviation between the mobile robot’s current magnetic heading direction and the one stored on the map, and then use this deviation to improve the localization using the scan matching method. The experimental results show that the mobile robot can perform much stronger localization during autonomous navigation.

Beach pollution, particularly the accumulation of garbage on seashores in Japan, poses a significant environmental threat. To tackle this persistent issue, beach cleaning activities are necessary. However, beach cleaning is not always scheduled regularly. It depends on the availability of volunteers, resulting in situations where they struggle to clean the beach due to excessive garbage or where there is no garbage left to clean, as the beach was cleaned a short while earlier. Hence, this paper proposes an unmanned aerial vehicle drone-based monitoring system powered with a garbage detection deep learning model. This system also integrates drone technology with a beach-cleaning ground robot to create an efficient cleanup system by planning the route for garbage collection. The research focuses on detecting garbage, locating garbage in the real world, clustering detected locations, and planning paths for ground robots. The system utilizes real-time garbage detection with the YOLOv8 model and georeferencing techniques to map garbage locations accurately. The project also employs hierarchical density-based spatial clustering of application with noise (HDBSCAN) and simulated annealing for optimal route planning. Experiments conducted in simulated and real-world environments, including Hokuto no Mizukumi Seaside Park and the Kyushu Institute of Technology Sports Ground, assessed the system’s accuracy. The results revealed a high detection accuracy of 98.33% in simulations, with an average root mean square error (RMSE) offset error of 0.25 m. In contrast, real-world conditions presented more challenges, resulting in a lower accuracy of 84.26% and an average RMSE of 1.96 m.

Recent advances in connected-vehicle technologies have enabled the large-scale collection of driving data, facilitating the deployment of data-driven control schemes. Although these methods offer advantages by eliminating the need for explicit modeling, they also introduce vulnerabilities due to their reliance on stored data. This study investigates a class of targeted data poisoning attacks on fictitious reference iterative tuning, a widely used data-driven controller tuning approach. We present a method that allows an adversary to influence closed-loop dynamics by manipulating the training data so that the resulting controller behavior matches a maliciously defined reference response. This strategy differs from conventional poisoning attacks, which aim only to the degrade control performance. Instead, it enables deliberate alteration of control characteristics such as overshoot and convergence time. The proposed attack is formulated as a constrained optimization problem under bounded tampering signals. Through a numerical study involving adaptive cruise control with stop functionality, we show that minor data modifications, indistinguishable from sensor noise, can cause significant degradation in control behavior. These findings highlight the need for robust security mechanisms in data-driven control implementation.

This paper is dedicated to the realization of an aircraft autopilot. Therefore, the authors focus on landing flights, which are considered the most difficult due to the physical contact between the aircraft and the runway. During landing, the aircraft descends toward the runway and momentarily maintains a constant altitude just before touchdown to mitigate the impact. The former scenario is referred to as “descending,” while the latter is known as the “landing flare.” The pilot performs different maneuvers for the aircraft, lowering the nose in the descending scenario and raising it in the flaring scenario. In this paper, therefore, we propose an autopilot specifically for unmanned landing flights. The autopilot is composed of a scenario classifier and two flight controllers, selected based on the classified flight scenario. This enables the aircraft to perform appropriate maneuvers autonomously depending on the scenario. Since only images from the cockpit are used as inputs, both the scenario classifier and the flight controllers are developed using convolutional neural networks. In the experiments, the proposed autopilot is applied to an aircraft in a flight simulator, and its effectiveness for landing flights is evaluated based on the results.

Recently, several researchers investigated mobile robots. However, when transporting objects in a complex and confined environment, robots that cannot move in any direction and cannot perform a super pivot turn have a limited range of motion. Therefore, omnidirectional mobile robots have been extensively studied. Mobile robots for transporting objects are often used in factories, public facilities, and restaurants. To shorten transportation time and avoid obstacles, mobile robots sometimes rapidly accelerate, decelerate, or traverse steps, which may cause vibrations and damage to transport objects. Transporting a person may give the person a sense of unease or discomfort. Therefore, vibration control is necessary. This study proposes the manual operation of an omnidirectional mobile robot to perform vibration control of a spherical pendulum, which is a tentative transport object. A notch filter and an optimal servo system were used for vibration control of the pendulum. The vibration of the pendulum generated during operation was reduced using a notch filter, and the vibration of the pendulum excited by the disturbances was suppressed by feedback control using the optimal servo system. The control gain of the optimal servo system was reasonably determined using a genetic algorithm based on a quantitative evaluation of the operability and damping performance. The effectiveness of the control system was verified through simulations and experiments.

This paper presents the development and evaluation of an omnidirectional mobile vehicle using a novel active omni wheel (AOW) mechanism. An AOW is a wheel mechanism capable of moving in an arbitrary direction by driving the wheel’s main body and outer rollers. The previous design features barrel-shaped outer rollers, which enables a seamless circumference and reduces vibration. However, it experiences slippage owing to friction-based transmission. To address this issue, a new AOW design incorporating positive drive mechanisms that utilize toothed belts is proposed to ensure non-slipping transmission and enhance the movement accuracy. The dynamic performance of a four-wheeled vehicle equipped with two AOWs and two passive omni wheels is evaluated by comparing the two configurations. The line-symmetric wheel layout with its center of gravity (COG) positioned near the AOWs exhibits superior performance in terms of maximum acceleration without wheel slippage. The point-symmetric wheel layout demonstrates a marginally lower performance but remains effective under varying COG positions. A prototype vehicle is developed and tested, verifying the effectiveness of the AOW in achieving omnidirectional movement. Furthermore, a control system is designed to manage actuation redundancy, which can result in load concentration on a single wheel. The proposed controller utilizes a PI control method to provide feedback based on the deviation of the wheel torque from its permissible value. The experimental results validate the capability of the proposed controller to ensure a balanced torque distribution between the AOWs.

In cooperative transportation, multiple robots share work that is difficult to perform using a single robot. This transformation enables a flexible combination of robots to transport objects, enabling efficient operation according to the situation. In recent years, the cooperative transportation of objects has been studied using formation-change algorithms with reinforcement learning. Although individual tasks, such as transport or formation change, have been studied, the coordination of all tasks in cooperative transport and control has not been discussed. In this paper, a formation-control system using a state machine is proposed for transportation tasks in a complex environment. First, reinforcement learning algorithms specialized for multiple agents were used to change the formation. As precise environmental sensing in the vicinity of a formation is required for cooperative transport, an integrated sensing system that shares omnidirectional camera and light detection and ranging (LiDAR) sensor information with all the transport robots was constructed. Next, the formation was controlled using a state machine with an integrated virtual LiDAR sensor. Finally, two scenarios with multiple robots were demonstrated to evaluate the effectiveness of the proposed system.

This study proposes a hierarchical coverage control method customized for heterogeneous multi-agent systems. In the proposed framework, agents are divided into leaders and followers. The leaders perform computationally intensive centroidal Voronoi control, whereas the followers track their assigned leaders using lightweight local rules. This division of roles addresses the challenge of the high computational demand of conventional coverage algorithms, thus making the method suitable for systems with limited onboard processing capabilities. Simulation results show that the proposed method significantly reduces computational load while achieving similar or superior coverage performance compared with conventional methods, even under non-uniform spatial importance distributions. The results demonstrate the feasibility of scalable and efficient coverage control for heterogeneous agents and indicate its applicability to real-world scenarios such as environmental monitoring and disaster response.

Swarm robotic systems consist of a large number of distributed autonomous robots that coordinate their actions to accomplish diverse tasks beyond the capabilities of a single robot. These systems have recently been considered for deployment in disaster scenarios, where communication is often unstable, making it necessary to achieve adaptive cooperative behavior without relying on explicit communication between robots. In the context of multi-robot systems—including swarm robotic systems—some studies have explored approaches utilizing large language models (LLMs) or other learning-based methods, but few have proposed systems that enable communication-free coordination. In this paper, we propose a system incorporating a novel method that combines high-level decision-making via LLM-based policy selection—guided by questionnaire-style prompts—with low-level control using multiple MARL-trained policies. We consider a complex task scenario in which robots search for a target object and transport it to a designated destination. To evaluate the method, we define implicit consensus as a condition in which a robot selects the same policy as its nearby robots without any explicit communication. The effectiveness of the proposed method is demonstrated through simulated task execution, with particular emphasis on implicit consensus as a key evaluation metric.

To make advanced driver assistance systems (ADASs) more accessible, it is essential to ensure their response is more approximated to the driver’s operating behavior. In this study, we proposed a method to adapt the ADAS to the driver’s behavior with high accuracy using an extreme learning machine (ELM) model, which enabled fast learning of the vehicle response. The learning object was the driver’s vehicle velocity control during deceleration to predict the future velocity. The future velocity could be used for model predictive control (MPC) of the reference velocity to achieve the desired vehicle behavior and guarantee safety and efficiency. The proposed ELM model consisted of a series of serially connected velocity predictors that covered multiple time step horizons for MPC. We developed a system where the ELM model learned the driver’s deceleration trajectory and the effectiveness of this system was assessed by applying it to an actual test vehicle. The results showed that the proposed ELM model could predict the velocity in real-time.

Control of pedestrian flows has various applications such as fire evacuation. Mobile robots play an important role in control of pedestrian flows. In this paper, a switching control method of pedestrian flows is proposed toward using of mobile robots in future. As a mathematical model for pedestrian flows, we adopt the discrete Hughes model, which is a kind of meso-scale models. In the discrete Hughes model, the space is represented as an undirected graph, and pedestrian flows are characterized by changes in density at each vertex. In the proposed method, pedestrian flows are controlled by dynamically switching the graph structure. For each graph structure, a safety penalty is assigned. The optimal switching times that minimize the penalty are determined under a constraint on the exit rate defined in this paper. Through a numerical example, we demonstrate the effectiveness of the proposed method in evacuation guidance.

As autonomous mobile robots (AMRs) are increasingly introduced into diverse environments, it is crucial that they interact smoothly and naturally with nearby pedestrians. To allow this, AMRs are expected to communicate and behave in a manner similar to interactions between humans. This study addresses pedestrian-pedestrian passing behavior and presents a novel quantitative index, termed the inducibility measure, that captures the extent to which a person’s actions can affect the behavior of those around them. Two types of inducibility measures are proposed: one derived from the sensitivity of the decision-making process, and the other from the controllability Gramian of a state-space representation of a closed-loop system representing interactive behavior. These measures were analyzed using a mathematical model of pedestrian behavior developed from actual observational data on pedestrian interactions. The proposed indices are intended to support the design and evaluation of AMR behavior, particularly in scenarios involving close interactions with humans.

In aging societies, soft multi-degree-of-freedom actuators are required for power-assist devices and welfare robots. Therefore, we previously developed a pneumatic artificial rubber muscle that uses shape-memory polymer (SMP) sheets with embedded electrical heating wires, which utilize the large difference in the elastic modulus below and above the glass transition temperature, shape fixity, and shape recovery of SMPs. Moreover, we increased the bending motion range of the actuators by segmenting the SMP sheet with embedded nichrome wires. We also improved the fabrication repeatability and shape recovery of the SMP sheets by using a shape-memory alloy (SMA) electric heating wire. In this study, we combine our previous improvements and propose a segmented shape-memory composite (SMC) sheet that consists of segmented SMP sheets with embedded SMA wires. We compare the shape, thermal images, and mechanical properties of the previously developed SMP sheets and the proposed segmented SMC sheets. The mechanical properties of the sample sheets are evaluated using a shape fixity and recovery test, a bending test, and a tensile test on prototypes. We also evaluate the motion of artificial muscles to which SMP sheets with embedded nichrome or SMA wires were attached using an isometric test and bending angle measurements. The experimental results confirm that the segmentation of the sheets improved muscle performance. Moreover, the use of the SMA wire improved fabrication repeatability and shape recovery, confirming that we successfully combined our previous improvements for SMP sheets with embedded heating wires.

This paper proposes a human-inspired navigation method that enables nonholonomic mobile robots to traverse dynamic pedestrian flows safely and efficiently. Conventional methods primarily focus on static environments or avoiding individual pedestrians; however, limited attempts have been made to validate robot crossing behavior experimentally within actual crowd scenarios. Therefore, this study implemented crossing navigation based on observed human behavior, specifically positioning robots behind pedestrians during crossing maneuvers. The algorithm developed herein identifies the optimal crossing points within dynamic pedestrian flows, considering robot nonholonomic constraints to ensure safety and efficiency. Additionally, potential field methods generate robot trajectories toward the identified crossing points, and pure pursuit control facilitates smooth trajectory tracking. Multiple simulations confirmed significant reductions in arrival time and path length compared with the conventional social-force-based method under uniform and nonuniform pedestrian flow conditions. Furthermore, pedestrian disturbances decreased, stabilizing the average walking velocities. In real-world experiments conducted in a 6 m × 6 m environment, robots successfully traversed pedestrian gaps without disrupting pedestrian flow. Moreover, pedestrians voluntarily yielded paths to the robot, indicating the importance of incorporating human social behaviors into robotic navigation planning. Thus, multiple simulation and experimental results demonstrated that the proposed method effectively balances safety and efficiency in robotic path planning through crowds.

In the design of robotic manipulators, achieving dexterity within a large workspace along with structural lightness remains a significant challenge. Conventional industrial robots, including serial and parallel robots, suffer from a trade-off between weight and workspace dexterity. In contrast, cable-driven parallel robots (CDPRs) offer excellent lightweight performance and a large workspace. However, their applicability is limited because their workspaces are constrained by surrounding frames. This paper is related to a 6-DOF manipulator that integrates an articulated manipulator with a CDPR. The end effector is a platform whose orientation is directly controlled by four cables, enabling 6-DOF motion with a lightweight structure. The manufactured prototype and architecture, as well as mechanisms involved in the efficient transmission of cable tensile forces, are detailed. The forward kinematics is analyzed, and the numerical solution using the Newton–Raphson method is reviewed. Simulations are conducted to validate the solution and confirm its feasibility. Furthermore, a position control method that incorporates platform statics is introduced. Experimental results confirm the trajectory tracking performance in both translational and orientational motions.

This paper presents a fully onboard control and planning system for indoor quadrotor navigation, leveraging modular control and lightweight motion planning. Autonomous operation of unmanned aerial vehicles (UAVs) in indoor environments, such as inspection, monitoring, and mapping, faces challenges due to the absence of global navigation satellite system (GNSS) and limited onboard computational resources. Existing methods often rely on external localization systems or high-specification CPUs, restricting deployment on compact, low-power UAVs. To address this, we develop a hierarchical control architecture comprising motor, attitude, velocity, and position controllers, all implemented with full-state feedback. Paired with a custom lightweight motion planning algorithm, the system operates efficiently on resource-constrained CPUs such as the Raspberry Pi. This approach enables affordable, low-power UAVs to achieve autonomous indoor navigation without external infrastructure. The results show stable flight performance and computational efficiency, validating suitability for GNSS-denied environments.

This paper investigated the dynamics and control of an underactuated rimless wheel (RW) with semicircular feet driven by two pendulum-like arms. We focused on the entrainment property of the RW walking to the periodic oscillations of the arms and analyzed how different arm coordination patterns, i.e., in-phase and anti-phase swingings, affect the walking stability and performance of the system. A mathematical model was developed that captured the continuous dynamics, collision events, and control method for this hybrid mechanical system. Numerical simulations demonstrated that the anti-phase arm swinging, similar to human walking, provided superior performance compared to the in-phase swinging. Specifically, the anti-phase coordination yielded a wider range of entrainment, as well as a wider range of stable walking, across various control parameters including the forcing frequency and the arm weight. Through rigorous Poincaré map stability analysis, we further quantified the system’s robustness to perturbations, revealing that anti-phase arm swinging maintains stability even with increased arm mass, while in-phase swinging becomes unstable. The stable walking of the anti-phase coordination was supported by the opposing movements of the left and right arms, which effectively counterbalance the forward and backward momentums to suppress the ground reaction forces. These results may provide design principles for controlling underactuated walking robots with arms.

Multi-robot systems are expected to underpin the future of automated infrastructure in various fields, including logistics and transportation. Despite the importance of multi-robot research, operating a large number of physical robots presents nontrivial challenges that exceed those encountered in simulation environments. These include communication saturation, the difficulties in real-time localization, detection of abnormal robot behavior, and the need for user-friendly interfaces to allow for effective state management. Collectively, they create a significant gap in the assumptions between lab-scale studies and deployable technologies in this field. Therefore, we developed a versatile testbed for multi-robot studies that can accommodate up to 100 differential-drive robots, called the 100-Mouse System. Our hallmark is scalability, while maintaining flexibility to allow researchers to test their ideas with actual robot fleets. This is a result of deliberately and integrally designed modules from both software and hardware perspectives, as well as a user interface, which is essential in large-scale experiments. As an application of this platform, we present a pattern formation task in a centralized manner, demonstrating the ability of the system to effectively accommodate large robot fleets.

A continuous, non-contact, and non-fixed observation of a specific point of a free-swimming fish body is effective for monitoring the health of fish (e.g., measuring the heartbeat). However, real-time high-resolution imaging is difficult because of the wide range of fish movements, including translation and rotation. In addition, the inter-individual crossing of multiple fish hinders a stable continuous observation of an individual for an extended period. We propose a system that enables the high-speed tracking of a sparse school of fish using elliptical self-windowing to address inter-individual crossing. The system also enables quasi-static imaging near the head of free-swimming fish using high-speed optical tracking and correlation-integrated elliptical self-windowing. Evaluation experiments showed that ellipse self-windowing is faster than 1 ms, including occlusion recovery in offline videos. We also demonstrated the sufficient sharpness of the high-speed optical tracking videos. The calculation of the normalization parameter for the video was fast (within 2 ms), had sufficiently low vibration, and was robust against inter-individual crossing.

In congested environments, achieving safe and efficient robot navigation without disturbing crowd flow is a significant challenge. Previous studies have not sufficiently validated motion control algorithms that account for the dynamic characteristics of pedestrian flows, leading to degraded navigation performance. In this paper, a merging and following control algorithm based on the real-time understanding of pedestrian flow dynamics is proposed to address this issue. The advanced recognition of pedestrian flows is enabled by the use of an overhead-view RGB camera and 2D-LiDAR, through which crowd dynamics are accurately captured. Based on the acquired dynamic flow information, merging motions that preserve crowd movement and following motions adapted to nonuniform flows are generated. The proposed system is evaluated in both nonuniform flow and separate flow scenes, which represent general crowd behavior. Multiple simulations and real-world experiments have demonstrated that the proposed system enhances navigation performance compared with conventional systems while ensuring human safety and comfort. In particular, reductions in travel time and path length, decreases in average acceleration and collision count, and the minimization of social force on pedestrians are achieved. These validation results indicate that the system enables robots to navigate crowds in a human-like manner, thereby supporting its potential deployment in practical environments.

Pneumatic artificial muscles (PAMs) are soft actuators that generate tension via contraction when supplied with compressed air. Although PAMs have been widely used in robots mimicking endoskeletal organisms, recent advancements in millimeter-scale thin PAMs have enabled a more precise replication of complex musculoskeletal systems. In contrast, exoskeletal organisms, such as crustaceans and insects actuate their joints using pennate muscles embedded within their exoskeletons. However, integrating actuators into the exoskeletons of exoskeletal-inspired robots is challenging owing to spatial constraints. Consequently, wire and servo-driven mechanisms are predominantly employed, and studies on exoskeletal robots incorporating the muscle-apodeme structures of exoskeletal organisms remain largely unexplored. To address this gap, this paper presents the development of a crab-inspired robotic walking leg featuring an exoskeletal structure with embedded pneumatic artificial pennate muscles, modeled after the muscle-apodeme structures of snow crabs. The experimental evaluations demonstrated that the robot successfully performed joint opening and closing motions, achieving a range of motion comparable to that of a snow crab.

An omnidirectional mobile robot that consists of two differential drive wheel modules and two links is proposed. Free-rotating joints are used to connect the links with each other and the wheel modules with both ends of the connected links. A caster is attached to the bottom side of the joint that connects the links. A ball caster is also attached to the bottom side of the steering axis of each wheel module. The robot has four driving wheels, two for each module. The robot utilizes the four degrees of freedom (DOFs) of these wheels to control three DOFs for omnidirectional locomotion and one DOF for the angle between the two links. The robot can change its form by changing this angle; this modifies the stable region of the robot. The robot prevents itself from falling down a slope by extending the stable region downward on the slope. As the robot uses regular rubber tire wheels, it has a large load capacity and high stability on irregular road surfaces. A robot with a depth camera is developed. The movement of the robot toward a goal detected using this camera is controlled. Experimental results show that the robot can move in any direction on a horizontal plane.

This study proposes a novel stair recognition method that integrates a monocular camera and a laser to improve the safety of the stair-climbing function in an omnidirectional autonomous electric wheelchair equipped with three Mecanum wheels mounted on a single axle. We performed three-dimensional stair measurements, estimated the angle of descent using a camera and laser, and built an automatic stair angle adjustment function into the wheelchair. The proposed method uses coordinate points to detect the staircase structure in three dimensions (x,y,z), performs distance transformation to achieve high-accuracy three-dimensional distance estimation, and provides a detailed visualization of the staircase geometry. Estimating the descent angle from the obtained 3D data yielded a maximum error of 2.72° and an average error of 1.04°, demonstrating higher accuracy than a stereo camera. Furthermore, the automatic stair angle adjustment function of the proposed wheelchair was validated, and an algorithm was developed to automatically maintain the wheelchair’s horizontal orientation based on the acquired stair angle. The experimental results confirmed that the proposed method can accurately adjust in real time with varying staircase angles, significantly improving the safety of the stair-climbing function. In addition, by applying this method to an autonomous mobile robot, it can detect obstacles and recognize staircase structures in the absence of ambient illumination, allowing for autonomous operation while analyzing its environment in three dimensions.

In recent years, the number of robots that interact with humans, such as guidance robots, has been increasing in commercial and public facilities. In educational-support robots have attracted attention in educational institutions owing to their effectiveness in supporting learning. Additionally, initiatives to advance the widespread use of robots through networked communication have led to the development, the Robot Service Network Protocol (RSNP), which is designed exclusively for robots. RSNP offers the advantages of secure communication with authentication and easy implementation across a wide variety of robots in a function-oriented library that accounts for multiple robot use conditions. In this study, we networked educational-support robots using RSNP and verified their practicality through load verification. Educational-support robots require a large amount of data to communicate during a single learning session, such as the facial expressions and grades of the learners, and secure and stable communication is required. In our experiments, we compared the data communication processing between RSNP and Hypertext Transfer Protocol (HTTP). Additionally, we conducted load testing of RSNP under multiple conditions that assume high-frequency communication processing. The experimental results show that RSNP achieves a processing speed comparable to that of HTTP while exerting minimal impact on overall processing performance. This study demonstrates that RSNP communication is more effective than HTTP for educational-support robots.