The network traffic prediction is essential for network operators in cost-effective link capacity planning. This paper proposes a method to predict traffic statistical upper limits (TSUL) as the required link capacity for a planning period of several months. Accounting for the large and non-stationary traffic fluctuations due to changes in flow numbers and bandwidth needs during the planning period is crucial. To adapt these fluctuations, we presents a network traffic prediction method for link capacity planning with two novelties. First, it uses flow features such as the number of flows and bandwidth requirements to adapt TSUL prediction to changing factors. Second, it utilizes a Supervised Variational Auto-Encoder (SVAE) model to estimate the traffic probability density function given flow features, and the predicted TSUL is derived from the upper limit of its confidence interval. In experiments using traffic data with large and non-stationary fluctuations, we evaluate the prediction accuracy of the proposed method in comparison with a linear probability density function estimation method (Linear) and the time-series prediction model (Long Short-Term Memory: LSTM and Temporal Fusion Transformer: TFT). Experimental results show that the proposed method is superior to Linear and LSTM in terms of TSUL root mean square error and log likelihood, and is inferior to TFT in terms of prediction accuracy but superior in terms of computation time. The experimental results suggest that the proposed method is suitable for link capacity planning targeting traffic with large and non-stationary fluctuations, considering not only the prediction accuracy but also the computation costs.

Factories have strict latency requirements and a large number of IoT devices, both of which are challenges when implementing software-based motion control systems. This paper proposes a new packet sequencing method for real-time level control of a large number of IoT devices with latency of better than 10 ms via the network by suppressing the congestion and jitter at the physical interface. The proposed method can guarantee real-time services of multiple devices from a single server by controlling the timing of transmission and reception. We develop a new driver for real-time industrial protocols and verify the effectiveness of the proposed method. An experiment confirms real-time response by successful controlling a motor with control cycle of 8 ms. Also, the proposed method is confirmed to connect 100 IoT devices with less jitter and packet loss compared to the conventional method.

In factory networks, various network protocols are used to meet strict requirements for accuracy and latency. The controllers that use these protocols are built with dedicated equipment, making it difficult to mix them with other protocols. Moreover, these controllers often rely on specialized software, complicating the integration of innovative technologies like AI. In this paper, we propose a versatile control system by implementing one of the industrial protocols, EtherCAT, on a general-purpose server. This system can control multiple protocols, allowing new protocols to be added without changing existing equipment. Experiments demonstrated that the system reliably operates multiple devices in a daisy-chain configuration, a characteristic feature of EtherCAT.

With the increasing demand for diverse services in access networks, ranging from legacy services to next-generation applications, network function virtualization (NFV) is being applied to access networks to enable agile service creation and flexible function deployment. While the virtualization of upper-layer management and control functions of optical line terminals (OLTs) has progressed apace, maximum flexibility by extending virtualization to lower-layer functions, including media access control (MAC) and physical-layer (PHY) functions, has been demonstrated using graphic processing units (GPUs). With improved scalability, fully virtualized OLTs have been shown to consume 31.8 W of power per OLT, which is still high, and no work focus on the realization of efficient virtual OLTs. In this paper, we propose a novel energy-aware GPU load balancing method for fully virtualized OLTs, and evaluate it on a multi-node supercomputer by running significant numbers of OLTs. The proposed method reduces the power consumption of compute nodes by minimizing the number of GPUs used for OLT deployment to match the traffic volume. In an evaluation, we run 320 OLTs on a supercomputer with multiple nodes having multiple GPUs OLTs are allocated to GPUs as determined by the load balancer proposed herein. The results shows that, when traffic volume is small, the proposed method can reduce the power consumption by approximately 98 % while maintaining throughput performance compared to the conventional round-robin method, and can hold consumption to 0.594 W per OLT.

In order to improve the flexibility and compatibility of industrial networks, there is ongoing research on the software-based implementation of industrial networks. The coexistence of multiple industrial protocols on the same industrial network is a challenge due to differences in QoS requirements and scheduling algorithms. In this study, we propose a method for the coexistence of multiple industrial protocols by introducing a protocol management application that manages the sending and receiving of packets in control from general-purpose computing resources. The proposed method is implemented and tested in a robot arm-based experiment. The proposed method achieved simultaneous operation of the robot arm and motor using different protocols. The proposed method could complete to control in 6.6 s even in a high-network-load environment, compared to the benchmark method, which took 256 s. In addition, a protocol switching experiment confirms that control could be continued without interruption even when the protocol was switched.

We experimentally demonstrate real-time end-to-end (E2E) uplink video streaming with our proposed jitter reduction technique for the use case of remote operation. One challenge is that the uplink video quality for remote operation is degraded by jitter due to the radio transmission scheme, such as time division duplexing (TDD), retransmission of lost data units, and data unit concatenation. The proposed technique executes traffic shaping at a router behind a next generation node B (gNB) in a mobile network by using the traffic information obtained from the gNB to calculate the optimal shaping rate. To evaluate the application-level performance of the jitter reduction technique, we carried out E2E uplink video streaming experiments using actual user equipment (UE), gNB, core node, and video server, including actual radio transmission and traffic shaping with the jitter reduction technique at an actual router. Experimental results of evaluating video frame intervals with real-time transport protocol (RTP) header information show that the ratio of video frame intervals that fell within ±5% of the ideal video frame interval value improved from 8.8% without shaping to 20.8% with shaping. These results experimentally demonstrate that jitter due to radio transmission actually degrades video quality in terms of video frame intervals and that the proposed jitter reduction technique effectively reduces it.

IEEE802.1Qbv time-aware shaper (TAS) has attracted attention as a technology to realize low-delay communication in various fields. Especially in industrial networks, the demand for low-latency communication between remote areas is increasing, and the realization of a service provider network (SPNW) using TAS is expected. Since SPNW frequently adds and removes users, increment scheduling is suitable for creating gate control lists (GCL) that control frame transmission for each node. The bandwidth utilization efficiency is worse when the order of scheduling cannot be controlled like in SPNW than when the order of scheduling is ideal. To cope with such problems, we focused on the difference in scheduling difficulty depending on the allowable jitter amount. This paper proposes a rescheduling scheme that enables scheduling of flows with small allowable jitter (high scheduling difficulty) by temporarily opening schedules of flows with large allowable jitter (low scheduling difficulty). The rescheduling scheme improves bandwidth utilization by up to 22.9%.

IEEE802.1Qbv (TAS: Time Aware Shaper) has been attracting attention as a technology to realize low-latency communications for industrial networks (NWs) and wireless networks. TAS achieves low latency by designing a transmission schedule called a Gate Control List (GCL) from NW information such as propagation delay and link speed and transmission information such as transmission period and frame size. The demand for low-latency communications between distant sites is increasing, and TAS is expected to be realized in Service Provider Network (SPNW). Applying TAS to SPNWs makes it difficult to design GCLs because GCLs need to be designed for all nodes that compose the large NW. Especially, when two or more transmissions with specific periods are multiplexed, the scheduling success rate deteriorates due to schedule collisions. In this paper, we aim to increase the success rate of scheduling by allowing for jitter in transmission. Experimental results show that the proposed method is able to schedule approximately 2.3 times more users with 5 μs of jitter tolerance than the method that does not tolerate jitter when multiplexing traffic with various periodicities. In addition, we found that by managing the jitter across all the nodes through which the transmission passes, the proposed method was able to suppress the jitter to about 13% on average compare to the method in which the jitter is managed independently at each node.

Because factory networks have strict requirements in terms of accuracy and latency, they are implemented using dedicated network protocols. However, these protocols tie the factory to specific equipment vendors, making it difficult to introduce to other protocols. In this paper, we propose a novel architecture that softwarizes the conventional industrial Ethernet protocol to replicate the functions of conventional industrial machines on a general-purpose server. Moreover, we first demonstrate the motion control from a softwarized control system using industrial Ethernet protocol and verify the effectiveness of the proposed architecture. Our proposal allows existing industrial protocols to be recreated in the software, making it possible to realize software-defined factory networks without changing the system configuration or the applications. The experiment confirms that the PROFINET software-based motion control system is implemented.

IEEE802.1Qbv (Time-aware Shaper: TAS) in IEEE802.1-Time-Sensitive Networking (TSN) has recently been applied within user networks (NWs) for realizing factory automation and remote control. TAS enables the coexistence of low-jitter traffic and best-effort traffic required in industrial NWs and enables deterministic control of communication latency. As mass customization matures, TAS is expected to extend its application coverage to not only user NWs but also wide-area NWs to enable real-time communication for remote control and cloud-based application use. TAS is TSN technology expected to be used for long-distance communications because it can guarantee maximum latency limits. However, designing a TAS schedule is known to be an NP-hard problem, which becomes more difficult as the number of forwarding hops increases. In this paper, we propose a method to solve the TAS design problem in large NWs by combining TAS design in smaller NWs; the NW is divided into several domains. We show that TAS solves the problem of large forwarding hop counts and improves schedulability when implemented in large NWs. This paper contributes to the realization of TAS in the wide-area NW, which has not been discussed so far.

As mass customization has matured, there is a growing expectation to extend the scope of IEEE802.1Qbv (Time-aware Shaper: TAS) to not only user network but also service provider networks (SPNW). To control communication latency in SPNW, a deviation in the transmission schedule of one user should not affect other users’ time-sensitive traffic. Hellmanns et al. [1] proposed a convergent TAS scheduling model. Kim et al. [2] proposed adding the concept of protection band to the TAS algorithm to send urgent event traffic immediately. The above related studies have not considered the resistance to the violation traffic when multiple users are accommodated in TAS NW, which we will discuss in this paper. First, we use the user identifiers to address the problem of other users' traffic interrupting the queue at different timing from the schedule by setting up exclusive queues and gates for each user. Second, we propose a combination of time slots that enable transmission of time-sensitive traffic and idle time that prohibits transmission. We take measures to prevent the transmission of delay-guaranteed traffic from colliding during transmission of other frames. A numerical simulation shows end-to-end jitter of SPNW is minimized by our proposed method.

Although demand has been increasing for accurate remote motion control of IoT devices, jitter can negatively impact control performance. Jitter buffering has been studied as a solution, but it is necessary to obtain delay values in real-time. In this study, we propose an architecture that uses collected network information to realize jitter buffering suitable for user-managed IoT device control applications in the All-Photonics Network. The architecture collects end-to-end delay values in real-time, implements jitter buffer based on the obtained values, and performs more detailed delay compensation control based on fixed delay values. This allows network conditions, which are normally out of user's control, to optimize the user's IoT device control application. We construct a basic motor-based experimental system around an FPGA and verify the effectiveness of the proposed architecture. An experiment shows that the proposed architecture holds jitter to any practical set value and achieves successful motor control within 0.62 s, whereas it cannot be controlled when there is no jitter buffer.

Machine learning-based network traffic prediction has been a hot research topic in recent works. The majority of recently developed prediction models have employed Long Short-Term Memory (LSTM) to predict the traffic at a few steps in the future from the most recent past traffic. However, for network provisioning to plan the capacity of a facility in a few days or months, it is necessary to derive its statistical upper limit from the traffic prediction results of a larger number of time steps. This paper investigates the network provisioning to estimate the conditional probability density function (PDF) of traffic given the future conditions instead of the traffic prediction for each step. Specifically, the network provisioning estimates the PDF of traffic given the future flow feature data and predicts the upper limit of the confidence interval of the estimated PDF as the facility capacity. This paper proposes two PDF estimation models are based on Supervised-Variational AutoEncoder (SVAE) and Student-t-SVAE (t-SVAE). Especially to deal with the biased traffic caused by the mixed flows with different features, such as elephant and mice flows, the t-SVAE uses a heavy-tailed Student-t distribution as the predictive traffic distribution. Our experimental result confirms that t-SVAE-based prediction can perform the high accuracy of the statistical upper limit prediction for provisioning networks where flows with different features are mixed.

Software Defined Networking (SDN) makes load balancing efficient and intelligent. However, fully deploying an SDN presents economical, organizational and technical challenges. Therefore, load balancing in hybrid SDN networks where the routing protocol parameters are designed through an SDN channel is one of the hottest topics in the field. In addition, symmetrically routing is a practical requirement for network operators since asymmetrically routed flows make the network too complex to manage. In this paper, we propose a load balancing method for symmetrically routed hybrid SDN networks that can handle existing distributed routing parameters, link cost and path selection. The proposed method optimizes link cost and path selection simultaneously under a symmetrically routed condition, while traditional methods optimize them individually. In a numerical simulation, the load balancing performance of the proposed method is better and more stable than any of the traditional methods. Furthermore, the proposed method has high versatility, and provides the best performance even in networks without distributed routing protocols.

A modular matrix converter (MMxC) is one of promising ac/ac converters for high power and high voltage application as a family of modular multilevel converters. For variable speed applications such as motor drive, a broad frequency range of operation is required. However, it has been pointed out that it is difficult for the MMxC to operate under the condition the output frequency is the same as the input. This paper focuses on a broad frequency range operation of the MMxC, and demonstrates the proposed control scheme using space vector modulation makes it possible to operate even at the same input and output frequency through numerical simulation and experiments. Moreover, detailed analysis on current distribution among converter arms are presented.

We propose a novel localization method for outdoor mobile robots using High Dynamic Range (HDR) vision technology. To obtain an HDR image, multiple images at different exposures is typically captured and combined. However, since mobile robots can be moving during a capture sequence, images cannot be fused easily. Instead, we generate a set of keypoints that incorporates those detected in each image. The position of the robot is estimated using the keypoint sets to match measured positions with a map. We conducted experimental comparisons of HDR and auto-exposure images, and our HDR method showed higher robustness and localization accuracy.

This paper presents a 3D laser scanner that can gaze at an arbitrary region. By modulating the secondary rotation speed of roundly swinging 3D laser scanner, it gazes at a specific region and measures with high density measurement points. The proposed method uses a secondary rotation motor to control the measurement point density and no extra motor is required. The 3D scanner can keep the field of view of and scan cycle time while gazing at any region. The system is evaluated on experimental targets for different scenarios using a mobile robot.

This paper provides valuable information about 3-D map construction using a tracked vehicle at Disaster City. Disaster City is a training facility for the Federal Emergency Management Agency (FEMA) in the United States. FEMA staff used our rescue robot Kenaf to conduct simulated emergency testing. Kenaf is a tele-operated tracked vehicle that has high mobility suitable for use in outside terrains and inside buildings. We equipped Kenaf with a 3-D laser scanner that can measure dense and wide angle 3-D shapes. We collected about 50 3-D mapping data sets during the tests. A part of these data sets can be accessed at our website. In this paper, we explain the Kenaf's tele-operation system, the 3-D mapping system, and the comments about the usability of the 3-D mapping from FEMA staff. In addition, we discuss about the limit of our mapping method on the basis of the trials at Disaster City. This information will be helpful for researchers in the robotics to improve their method of 3-D map construction and their tracked vehicles.

This paper proposes a new localization method for outdoor navigation using a stereo camera only. Vision-based navigation in outdoor environments is still challenging because of large illumination changes. To cope with various illumination conditions, we use 2D occupancy grid maps generated from 3D point clouds obtained by a stereo camera. Furthermore, we incorporate salient line segments extracted from the ground into the grid maps. This grid map building is not much affected by illumination conditions. On the grid maps, the robot poses are estimated using a particle filter that combines visual odometry and map-matching. Experimental results showed the effectiveness and robustness of the proposed method under various weather and illumination conditions.

The goal of rescue robotics is to extend the capabilities and to increase the safety of human rescue teams. During a rescue mission a mobile robot is deployed on a rescue site and is operated from a safe place by a human operator. The operator can not see the robot and the environment and a decision on the path selection is very complicated. Our goal is to provide a kind of automatic “pilot system” to propose an operator a good direction or several options to traverse the environment, taking into an account the robot's static and dynamic properties. To find a good path we need a special path search algorithm on debris and a proper definition of a search tree, which can ensure smooth exploration. While the main goal of the algorithm is to keep the robot maximally stable at every step of its path, in some cases we need the robot to lose its balance and to change a 3D orientation discontinuously. Losing balance on purpose is an important feature for safe climbing up and going down the debris and it is the central issue of this paper. Exhaustive simulations were used to structure and analyze data and experiments were used to verify our approach to removing unsuitable directions of the search from the search tree.

Tracked vehicles are frequently used as search-and-rescue robots for exploring disaster areas. To enhance their traversability on rough terrain, some are equipped with “active flippers.” However, manual control of such flippers also increases the operator's workload, particularly for teleoperation with limited camera views. To eliminate this tradeoff, we developed a shared autonomy system using an autonomous controller for flippers that is based on continuous three-dimensional terrain scanning. In our system, real-time terrain slices near the robot are obtained using three laser range sensors, and these are integrated to generate three-dimensional terrain information. In this paper, we introduce the autonomous controller for the flippers and validate the reliability of the shared autonomy system through experimental results on actual rough terrain.

This paper proposes a light-weight sensor platform that consists of gyro-assisted odometry and a 3D laser scanner for localization of human-scale robots. The gyro-assisted odometry provides highly accurate positioning only by dead-reckoning. The 3D laser scanner has a wide field of view and uniform measuring-point distribution. Robust and computationally inexpensive localization is implemented on the sensor platform using a particle filter on a 2D grid map generated by projecting 3D points on to the ground. The system uses small and low-cost sensors, and can be applied to a variety of mobile robots in human-scale environments. Outdoor navigation experiments were performed at the Tsukuba Challenge 2009, which is an open proving ground for human-scale robots. Our robot successfully navigated the assigned 1-km course in a fully autonomous mode multiple times.

We are developing robots for inspection of the under floor of houses. The developed robot system is reported in this paper. The mobile robot platform is based on a crawer type robot developed for rescue purposes. As sensors, a pan-tilt-zoom camera, top-view camera and SOKUIKI sensor are used. We constructed a test field and verified the performance of the robot.

Active flippers for tracked vehicles are very useful to improve traversability on uneven terrain. However it is widely known that control of flippers also increases the work-load for operators, particularly where the vehicle and the operator are far apart. To reduce the work-load, we aim to realize a sensor-based autonomous controller of flippers to enable a “semi-autonomous operation” of tracked vehicles. The “semi-autonomous operation” means that the only requirement for an operator is to indicate the robot’s direction. In this way, the robot is navigated autonomously through its sensors and actuators to surmount or avoid obstacles. In this research, two laser range sensors are used for terrain sensing, and gyro sensors are used for the measurement of the robot’s attitude. Based on such sensor system, we propose a strategy of simple sensor-based motion of active flippers for tracked vehicles to enable a semi-autonomous operation. In this paper, we introduce a strategy of motion of active flippers, and the stability analysis of tracked vehicles with active flippers. Finally, we report several experimental results to verify the validity of our approach.

Active flippers for tracked vehicles are very useful to improve traversability on uneven terrain. However it is widely known that control of flippers also increases the work-load for operators, particularly where the vehicle and the operator are far apart. To reduce the work-load, we aim to realize a sensor-based autonomous controller of flippers to enable a ‘semi-autonomous operation’ of tracked vehicles. In this research, two laser range sensors are used for terrain sensing, and gyro sensors are used for the measurement of the robot’s attitude. Based on such sensor system, we propose a strategy of simple sensor-based motion of active flippers for tracked vehicles to enable a semi-autonomous operation. In this paper, we introduce a strategy of motion of active flippers. Finally, we report several experimental results to verify the validity of our approach.

A rescue crawler robot having flipper arm has high ability to get over rough terrain, but it is hard to control its flippers in remote control. The authors aim at development of a semi-autonomous control system for the crawler robot. In the system, moving direction is specified by an operator at remote place. Joypad (Sony PS2 controller) is used as input devise. For increasing its stability and robustness about change of the environment, its flippers are controlled according to sensor informations which can be obtained in real-time. Concretely, the robot recognizes the environment (upward step or downward step) using its physical model, its postural information, flippers' contact and distance between the body and ground, and controls these flippers. In this movie, we show you performance of our proposed control method using "Aladdin".

A rescue crawler robot with flipper arms has high ability to get over rough terrain, but it is hard to control its flipper arms in remote control. The authors aim at development of a semi-autonomous control system for the solution. In this paper, the authors propose a sensor reflexive method that controls these flippers autonomously for getting over unknown steps. Our proposed method is effective in unknown and changeable environment. The authors applied the proposed method to Aladdin, and examined validity of these control rules in unknown environment.

A mobile robot can use the walls or the pillars to localize itself when it navigates in indoor environment. However, in open space or outdoor environment there are few easy detectable and stable objects that the robot can use. In this paper, we propose a method for the detection of braille blocks, which are originally used for visually handicapped people, for autonomous mobile robot navigation. To recognize the braille block, a CCD camera and a laser fan beam projector are used as sensor to detect bumps on road surface. This paper also presents the experimental results of braille block following using the sensor system to detect the braille block position and orientation. This experiment shows effectiveness of the sensor system for the braille block recognition and implies the possibility of braille block based mobile robot navigation.