In this study, we modified a commercially available small bipedal robot to be able to walk on the

. First, to prepare the experimental environment, we modified the biped robot and made a stand. Then, we checked the operation of the motors. After confirming the operation of the motor, we created a program to make the robot do the walking motion. Then, using the program, we created a walking motion and conducted a walking experiment. Now that we have confirmed that the biped robot can walk, we would like to conduct research to apply the information obtained from 3D spatial recognition to biped walking by installing a depth imager on the robot.

抄録全体を表示

In order for robots to operate safely, it is important to sense the environment. However, creating a sensing system from scratch is costly. To improve this situation, we use ROS2. In addition, we thought that a small mobile robot can move safely if it can grasp the space when sensing. Therefore, we use the Intel RealSense Depth Camera D435, which is a depth camera, and the

, which is small and easy to integrate with a depth camera. Therefore, in this paper, we use the latest version of ROS2 Foxy Fitzroy, and Intel RealSense Depth Camera D435, to a with 64bit version Ubuntu 20.04LTS installed, and measure the environment with the connected depth camera. We will measure the environment with the connected depth camera.

抄録全体を表示

In this research, we develop a small wheeled inverted pendulum robot, μ-Pentar, equipped with a

B. Due to the shortage of semiconductors in this year, we were unable to obtain an FPGA which was to control the motor driver and perform the signal processing for the encoder. Therefore, we investigated a method using the pigpio library instead of the FPGA. In this paper, the results of measure the response time of encoder counter function of pigpio library and the latency of serial communication line for IMU that measures the posture of the robots.

抄録全体を表示

A low-cost RTK-GNSS base station using a with a cellular sim card as a server was developed for precise positioning. A u-blox ZED-F9 P GNSS module connected to an antenna was connected to the for correcting RTK signals. An open source NTRIP 1.0 caster was utilised after which a tractor, fixed-wing UAV and multi-rotor UAVs could be connected at the same time with a fast time-to-first fix (TTFF). For accuracy comparison, it was found that positions calculated with RTK data from Memuro Base Station had a lower horizontal error of 7.2±4.5 mm compared to that of conventional subscription Nippon GPS Data Service which had a horizontal error of 9.2±5.75 mm.

抄録全体を表示

3 actuators concentrated leg mechanism around thigh axis was designed for quadruped dynamic walk. The Dynamixel XM430-W350-T, ROBOTIS Co. Ltd., was selected as the higher torque actuator unit instead of Dynamixel AX-12A. Leg length was determined from the catalogue torque of Dynamixel XM430-W350-T and necessary torque to stand up statically and accelerate dynamically. Authors adopted

as controller and developed an extension board for control. The trot gait walk was experimented.

抄録全体を表示

From the perspective of security in a living space, it is useful for a patrol robot to check and perform door closing, and this research focuses on the door closing task by a small robot. While small robot is easy to install in a living space, it has a problem that it cannot continue the task when closing a door that is large and heavy compared to its own body because of its small size. To solve this problem, the robot recognizes an open door by patrolling with an RGB-D camera and tries to hit the door without losing its posture. As a result, the door can be closed even if it is twice as heavy as it should be. The robot can also autonomously determine the path of door closing based on the presence or absence of the door closer, and performs a series of operations by removing the door stopper that serves as an obstacle with a manipulator. Finally, the system confirms whether the door is normally closed and takes appropriate action, and confirms and executes the door closing process by sequentially determining whether the door is open or closed.

抄録全体を表示

We propose an educational model that simulates a CubeSat for use in improving Japanese high school students’ recognition of the usefulness of science. This model allows high school students to directly relate the components of satellites to the physics concepts they are studying in class to improve their understanding of the relationship between space exploration and their classwork. We conducted a science workshop for junior high and high school students using an educational model simulating a CubeSat, consisting of a

and a Sense HAT (TopGunSat). The results of a post-workshop questionnaire suggested that the activity using TopGunSat was difficult but interesting for the participants. We also measured changes in the participants’ interest in science, mathematics, and programming between before and after the workshop. We identified a statistically significant improvement in programming related to daily life.

抄録全体を表示

Locomotive robots consist of wheeled, crawler type, and legged type. Dynamic walking can increase propelling velocity on the flat ground and get over the step. The Dynamixel AX12-A was used in the previous study to make the quadruped walk. The XM430-W350-T with high output torque was selected in this study. Leg length was limited from the torque of Dynamixel xm430-W350-T legs in statics analysis. Author tried to analyze the dynamics using the Lagrange equation to obtain the required torque. Author changed the robot hardware from Arduino to the more powerful

and developed an extension board for control.

抄録全体を表示

We are conducting research and development of mROS 2, that is an agent-less and lightweight runtime environment for ROS 2 node onto embedded devices. In this paper, we make the communication protocol stack of mROS 2 compliant with POSIX (Portable Operating System Interface). Since POSIX is a unified interface standard for operating systems, this work enables mROS 2 to operate easily onto Ubuntu OS, that is the default platform for ROS 2. We implemented the proposed method on Ubuntu 20.04 running on

B. Experimental result showed that our research outcome could improve communication performance than the native ROS 2 environment.

抄録全体を表示

This paper analyzes the impact of input sparsity and DFS/DVFS configurations for single-board computers on the execution time, power, and energy of each VGG16 layer as the first step towards efficient CNN inference on single-board computers. For this purpose, we first develop a power and execution time measurement environment and perform experiments using

and NVIDIA Jetson Nano. Our results show that clock frequency strongly correlates with execution time and power. Inversely, input sparsity has a weak correlation with execution time and power. Then, we show that a coarse-grained DVFS model can explain over 96% of the variations in the power of each VGG16 layer even when sets of clock frequency and voltage on the single-board computer are unavailable.

抄録全体を表示

Studies have been conducted on tidying up objects on the floor by mobile robots. Problems in the studies are i) limiting the number of objects due to the number of manipulators, ii) difficulty of activities in the space with a low ceiling, and iii) limiting the size of objects due to the size of the robot body. This study proposes a manipulator which attaches a bag to the end-effector of a parallel link mechanism attached a bag and a small mobile robot equipped with the manipulator. The bag has the advantages in which it can carry multiple objects, it is smaller when it is not used, and it can carry long objects. These advantages solve problems (i), (ii), and extend the limit of problem (iii) to long objects. The robot has three manipulators: one with a bag and two for auxiliary collection. The auxiliary manipulators are used to collect multiple objects, collect long objects, and return from turnover during transportation. For problem (i), it realizes tidying up multiple objects that the auxiliary manipulator grasps an object and drops it into a bag when other object is in a bag. For problem (ii), it is possible to continue the task in the space with a low ceiling by reducing the height of the robot when a bag is not used. For problem (iii), it is possible to tidy up a long object by pushing a long object against a surrounding wall and placing it in a bag. Finally, this study demonstrates tidying up multiple objects which includes collection, transportation, and discharge by image recognition．

抄録全体を表示

CYber PHysical overlay network over Internet Communication (CYPHONIC) has been proposed as a communication architecture that simultaneously achieves communication connectivity and mobility transparency in a mixed IPv4/IPv6 environment. Using CYPHONIC, applications running on mobile devices and IoT devices can realize end-to-end encrypted communication across an overlay network. However, if firewalls installed on the communication path between end nodes do not allow the CYPHONIC protocol, the overlay network cannot be constructed. This paper proposes CYPHONIC-over-QUIC, which integrates QUIC, a standardized general-purpose transport protocol designed for web communications, into CYPHONIC to provide end-to-end encrypted communications that can pass through firewalls and NATs. We implemented CYPHONIC-over-QUIC on two

s and Linux servers running on AWS EC2, and evaluated its communication performance using the actual Internet environment. As a result, we confirmed that the signaling process at the start of communication does not affect the application communication and that the throughput performance is equivalent to that of the conventional CYPHONIC.

抄録全体を表示

A practical issue of edge AI systems is that data distributions of trained dataset and deployed environment may differ due to noise and environmental changes over time. Such a phenomenon is known as a concept drift, and this gap degrades the performance of edge AI systems and may introduce system failures. To address this gap, retraining of neural network models triggered by concept drift detection is a practical approach. However, since available compute resources are strictly limited in edge devices, in this paper we propose a fully sequential concept drift detection method in cooperation with an on-device sequential learning technique of neural networks. In this case, both the neural network retraining and the proposed concept drift detection are done only by sequential computation to reduce computation cost and memory utilization. We use three datasets for experiments and compare the proposed approach with existing batch-based detection methods. It is also compared with a DNN-based approach without concept drift detection. The evaluation results of the proposed approach show that the proposed method is capable of detecting each of four concept drift types. The results also show that, while the accuracy is decreased by up to 0.9% compared to the existing batch-based detection methods, it decreases the memory size by 88.9%-96.4% and the execution time by 45.0%-87.6%. As a result, the combination of the neural network retraining and the proposed concept drift detection method is demonstrated on Raspberry Pi Pico that has 264kB memory.

抄録全体を表示

Considering the growth of the IoT network, there is a demand for a decentralized solution. Incorporating the blockchain technology will eliminate the challenges faced in centralized solutions, such as i) high infrastructure, ii) maintenance cost, iii) lack of transparency, iv) privacy, and v) data tampering. Blockchain-based IoT network allows businesses to access and share the IoT data within their organization without a central authority. Data in the blockchain are stored as blocks, which should be validated and added to the chain, for this consensus mechanism plays a significant role. However, existing methods are not designed for IoT applications and lack features like i) decentralization, ii) scalability, iii) throughput, iv) faster convergence, and v) network overhead. Moreover, current blockchain frameworks failed to support resource-constrained IoT applications. In this paper, we proposed a new consensus method (WoG) and a lightweight blockchain framework (iLEDGER), mainly for resource-constrained IoT applications in a permissioned environment. The proposed work is tested in an application that tracks the assets using IoT devices (

Pi

and RFID). Furthermore, the proposed consensus method is analyzed against benign failures, and performance parameters such as CPU usage, memory usage, throughput, transaction execution time, and block generation time are compared with state-of-the-art methods.

抄録全体を表示

A backdoor attack is a type of attack method inducing deep neural network (DNN) misclassification. An adversary mixes poison data, which consist of images tampered with adversarial marks at specific locations and of adversarial target classes, into a training dataset. The backdoor model classifies only images with adversarial marks into an adversarial target class and other images into the correct classes. However, the attack performance degrades sharply when the location of the adversarial marks is slightly shifted. An adversarial mark that induces the misclassification of a DNN is usually applied when a picture is taken, so the backdoor attack will have difficulty succeeding in the physical world because the adversarial mark position fluctuates. This paper proposes a new approach in which an adversarial mark is applied using fault injection on the mobile industry processor interface (MIPI) between an image sensor and the image recognition processor. Two independent attack drivers are electrically connected to the MIPI data lane in our attack system. While almost all image signals are transferred from the sensor to the processor without tampering by canceling the attack signal between the two drivers, the adversarial mark is injected into a given location of the image signal by activating the attack signal generated by the two attack drivers. In an experiment, the DNN was implemented on a

to classify MNIST handwritten images transferred from the image sensor over the MIPI. The adversarial mark successfully appeared in a specific small part of the MNIST images using our attack system. The success rate of the backdoor attack using this adversarial mark was 91%, which is much higher than the 18% rate achieved using conventional input image tampering.

抄録全体を表示

As robotic systems become increasingly integrated into plant phenotyping processes, the quality of images captured plays an increasingly crucial role in accurate data collection and analysis. Here we address the challenge of assessing and enhancing image quality in the context of plant phenotyping robot using a pan-tilt-zoom camera. We present a data-driven approach using machine learning, specifically the Random Forest classifier, to classify both blurred and sharp images. Our method involves feature extraction, data preprocessing, hyperparameter tuning, and cross-validation. The resulting model demonstrates promising performance as indicated by its accuracy, precision, recall, area under the curve (AUC), and feature importance analysis. Notably, our results support a highly accurate classifier, achieving a correct classification rate of 95% for sharp images and 92% for blurred ones, a receiver operating characteristic curve with an AUC of 0.93, and a precision-recall curve with an average precision of 0.91. Shapley Additive Explanations analysis identifies “edge density” and “mean gradient magnitude” as influential to the classifier, offering valuable insights for future feature engineering and model refinement. The classifier has a short inference time (2.8 s) on a

B computer, both improving the quality of captured images and automatically eliminating blurred images. By enhancing image quality assessment, this research improves data reliability and the overall effectiveness of plant phenotyping robots. We discuss the implications of these findings and their practical relevance and suggest directions for future research.

抄録全体を表示

This paper presents a deep network based on morphological filters for Gaussian denoising. The morphological filters can be applied with only addition, max, and min functions and require few computational resources. Therefore, the proposed network is suitable for implementation using a small microprocessor. Each layer of the proposed network consists of a top-hat transform, which extracts small peaks and valleys of noise components from the input image. Noise components are iteratively reduced in each layer by subtracting the noise components from the input image. In this paper, the extensions of opening and closing are introduced as linear combinations of the morphological filters for the top-hat transform of this deep network. Multiplications are only required for the linear combination of the morphological filters in the proposed network. Because almost all parameters of the network are structuring elements of the morphological filters, the feature maps and parameters can be represented in short bit-length integer form, which is suitable for implementation with single instructions, multiple data (SIMD) instructions. Denoising examples show that the proposed network obtains denoising results comparable to those of BM3D [1] without linear convolutions and with approximately one tenth the number of parameters of a full-scale deep convolutional neural network [2]. Moreover, the computational time of the proposed method using SIMD instructions of a microprocessor is also presented.

抄録全体を表示

The rapid economic development has fueled the demand for enhancing lifestyle and home aesthetics, leading to the growing popularity of leisure activities and home decoration. As a response, the ornamental fish industry has flourished, prompting fish enthusiasts to seek efficient ways to care for their fish. Smart fish box has emerged as popular solutions, offering features such as remote control and monitoring. Smart fish box incorporates machine vision and Internet of Things technologies, allowing users to remotely control lighting, water changing, feeding, and oxygen pump operations. Temperature sensors transmit data to a mobile app, enabling users to monitor and adjust water temperature. These boxes also features built-in cameras for real-time monitoring and send notifications when fish food is running low. This innovative design addresses several challenges in ornamental fish care. This paper presents the mechanical structure, control circuitry, and vision algorithm of the smart fish box. By utilizing collected data, a neural network is trained on the Raspberry Pi platform, successfully recognizing fish health status.

抄録全体を表示

自己位置推定等においてステレオカメラで画像取得を行う際、CPUのみでは時間差が生じてしまう。そこで、並列処理、並列I/Oが可能なFPGAを用いることでCPUのみで処理をするシステムに比べて効率的なシステムの構築が期待できる。再設計可能で低消費電力なFPGAの特徴を生かしたリアルタイム画像処理が可能なシステムの構築、システムの通信モジュールにおける通信効率の最適化、通信速度の測定を行い、十分なパフォーマンスを確認した。

抄録全体を表示