The objective of this paper is to predict the turbidity after flocculation from floc images using deep convolutional neural network(DCNN) with small flocculation plant. Our goal is to develop a system to control the water purification process using the predictive model. The following results were recognize from experiments using floc images from a small flocculation plant: 1) the DCNN model is able to recognize the image characteristics of the flock; 2) the prediction accuracy is around 0.10 for the teacher data and around 0.31 for the test data.

Parallel link mechanism is expected as a sort of mechanical device of machine tools and robots for realizing high positioning accuracy and high rigidity. For practical application, however, the problem of inverse kinematics has to be resolved on the position aberration occurred by machine difference, the bending caused by the collision of redundant drive system, and so on. To cope with the problem, we propose a method for defining inverse kinematics function which is represented by polynomial interpolation function tailored as obtaining physical correspondence with the posture of parallel link mechanism and the instruction of drive system. And we show the feasibility and effectiveness of the proposed method through a numerical experiment using theoretical inverse kinematics function and an application for practical parallel link mechanism.

Much research for motion generation using machine learning is keeping attention for adaptation to various environments. Imitation learning that can generate robot motions from human demonstrations is a good candidate. We conventionally showed the usefulness of bilateral control to collect motion data for imitation learning. However, image information has not be fully utilized in the method. In particular, the picking task, which has been widely studied as a task involving contact with objects, has not yet been achieved. In this study, we set a conveyor picking task as the target task and verify a learning method that combines ibilateral control based imitation learning and image processing.

Many robot control systems that require complex processing rely on external computing resources such as ROS. We believe that the usability of such a configuration is limited because the communication delay affects the control. Therefore, this paper standalone robot control system using hierarchical deep reinforcement learning.

In this paper, we apply the Actor-Critic method, which is one of the machine learning methods, to swing-up problem of an Acrobot. First, we train an agent to reward the energy of the first link of the robot. At this time, we use the model that limits the amplitude of the second link. We confirm by simulations that the energy can be increased from a swing-up motion to a giant swing motion by the agents trained for the several amplitudes of the second link.

This paper describes how to create a controller for continuous brachiation including aerial phase by reinforcement learning. There are no previous research that have achived continuous brachiation including aerial phase. Because of the need for efficient search and robust controller, we used reinforcement learning including curiousity and proposed a branch trimming method using a neural network. We evaluate the performance of learning using the proposed method on a simulator, and then control the actual machine using the learned policy.

In recent years, many systems and robots have been installed in artificial environments such as urban areas and factories.They are also operated in natural environments such as forests and waters.In such places and scenes, the arrangement of people and objects is constantly changing due to their moving, and in addition, there is a possibility that an unknown environment which engineers cannot preliminarily estimate would appear.In order for a plurality of robots to perform stable operations in such unstable environments, some cooperation between the robots must be important.Assuming that the whole is regarded as an autonomous decentralized system in this study, we examine what kind of interactions between multiple robots (agents) can be formed when reinforcement learning based on trial and error is individually applied to each robot (agent).

For autonomous mobile robots to act in society, it is important to predict the action of dynamic obstacles. In this paper, we use state representation learning, a form of deep learning, to obtain a predictive model from 2D-LiDAR and the robot’s action that considers the interaction between the robot and dynamic obstacles. This predicts the future for possible actions taken by the robot. The prediction results are compared with GRU, a form of representation learning. We also describe how to use the prediction results for conventional motion planning. As for as we know, there are few cases where prediction results are reflected in motion planning.

Parallel link mechanism has the potential of high rigidity and high positioning accuracy. Its control operation, however, is difficult to usage as mechanism for machine tools and robots because of machine difference and collision of driving torque. So direct teaching operation is expected in the practical usage of parallel mechanism as robot mechanism. Its mechanical characteristics have to be required for realizing direct teaching operation, however, it is not well known for parallel link mechanisms. In this paper we investigate the relation between the control force applied for the mechanism and the torque of drive systems in the purpose to realize direct teaching operation of 3R4 type spherical parallel link mechanism.

In this research, we propose an AGV with a simplified mechanism and structure using “Karakuri” technology. The developed AGV can be driven and steered by a DC motor. A pair of two cams that can change the traveling route by changing the it’s shape is adopted for the steering mechanism. From the experimental results, it was confirmed that it is possible to steer left and right with the cam mechanism. However, the measured turning radius on the left and right was different from expected radius, since the tension of the wire used for the power transmission of the steering mechanism is influenced.

In 2025, Osaka will host a major world exposition centered on the theme of ”Designing Future Society for Our Lives” where new technologies will be showed. This paper showcases our first prototype of a garbage collecting robot which took first place in a garbage collection tournament organized in the Nakano-Shima event in Osaka in December 2021. This first prototypes focuses on an easy to use remote control approach meant to be used when the automated robot fails to complete the task on its own. One of the main benefit of our design is the usage of an universal gripper for garbage manipulation. In this paper, we will describe the different parts of the robots as well as the design choices made for this first protoype.

Safer measures are required for work at height where humans are suspended from a rope in order to reduce the number of fatal and injury accident. In this study, we developed and tested a robot that can work at heights on the wall of a building in place of a human. We have constructed an under-constrained cable-driven window cleaning robot for portability. In order to solve the lack of pushing force of cleaning tools at the lower floor of the building, electrical ducted fans are installed to compensate for the pushing force.

In schools and other educational institutions, there are many opportunities to carry classroom chairs according to purpose. Some chairs are on casters, but educational institutions typically use stacking chairs or folding chairs. Although effective in terms of storage, it must be lifted by hand during transportation, increasing the burden on workers. Automation of baggage transport in warehouses has improved significantly, but little attention has been given to chair transport automation. Self-propelled chairs were developed, but despite the need, automatic transport of chairs without casters were not noted. In this study, we constructed an automatic chair movement system using an omnidirectional mobile robot. Recognize the chair by image processing and bring the robot closer. After the robot dives under the chair, the lifting device attached to the omnidirectional robot is used to lift the chair and transport it to the front of the human.

With the labor shortage, autonomous mobile robots are required to perform a variety of tasks. When we consider having an autonomous mobile robot perform tasks such as cleaning, maintenance, and inspection, it is assumed that the tasks will change depending on its own position. Therefore, it is difficult to perform the task effectively unless the robot knows its own position and posture with higher accuracy than in the case of simply moving. In this study, we are developing a robot that can move in all directions using a mecanum wheel. And also we are developping an autonomous movement system using a local path planning method that can determine the movement direction and posture independently to make effective use of the omni-directional movement mechanism, and a global path determination method that assumes that a non-expert person can set the path.

Due to the rapid aging of population, the most of agricultural work are done by farming machines. Tractor operation requires a high level of skill, and the constant posture of the tractor driver brings a heavy burden on the driver. On the other hand, the automatic tractors are usually expensive. They also have to buy ready-made products because they have automated internal systems. This research aims to automate the operation of tractors in order to reduce the burden on the people who operate tractor, and to develop a low-cost, add-on automatic driving system that can be used for tractor and other vehicles.

In this paper, we describe the development of a wheel for a mobile robot that polishes steel structures. Although there are various ways of moving and adsorbing a wall for a mobile robot, we adopted a method of moving by wheels and adsorbing by a permanent magnet because it is necessary to keep applying a constant pressure when polishing. A ring-shaped magnet is incorporated into the wheel to ensure constant polishing pressure with constant adsorption force while moving. We confirmed that the robot equipped with these wheels can move freely on flat walls, slopes, and ceilings. Two brushless motors are mounted in the driving system, and differential drive enables the robot to move straight and turn on the spot. In the prototype, commands are sent from a PC to operate forward, backward, and turning movements. In this paper, we report the results of testing the running performance of the prototype.

This paper describes a pointing gesture interface for an AMR used in factory transportation. The pointing gesture by a person is not always precise and contains ambiguity because of the person’s habit. Previous research on pointing gestures has focused on estimating the exact pointing position, which affects the accuracy of robot guidance. Our research assumes that human pointing is ambiguous. Our interface avoids estimating the 2D position on the floor from human pointing, and only estimates the pointing direction. By combining the recognized pointing direction with object recognition using RGB images, our system could guide the AMR to a specific position.

At logistic sites, the workers need frequently load luggage to and unload luggage from carts while bending over their lower back. It brings the burden on the lower back of the worker, which causes back pain. In this research, we developed a new type of electric powered lift cart that can automatically keep the package height on the cart table to a position by distance sensors, in which the height is the most comfortable for workers’ loading and unloading. Moreover, we developed a lift mechanism for the cart in which the driving force is constant so that using a comparatively low power motor is possible.

This paper describes the vibration suppression of a slung load which is carried by unmanned aerial vehicle (UAV), by tracking the vehicle to a vibration suppression trajectory. The trajectory is shaped in the frequency domain not to stimulate the normal mode of the UAV-load system. Since the frequency-domain design can explicitly handle constraints on acceleration and velocity of UAV, the resulting trajectory has time-suboptimality. The UAV-load system in a vertical plane is firstly modeled as DAEs and then the vibration suppression trajectory is designed using the natural frequency of load vibration and the vehicle constraints. Vibration suppression of the slung-load is verified in numerical simulations. The vibration control is finally implemented on ArduPilot, an open-source autopilot system.

In this paper, we tested a method of opening and passing through door for a mobile manipulator. Especially, we assumed the mobile manipulator constructed by commercially available robots; the arm robot is position controlled and the mobile robot automatically generates the path. We selected a method by Ott et al. as a promising method and implemented on a mobile manipulator. We revealed that the initial position affected to the force and moment acting on the tip of the arm robot and that it can provoke overload to fail the door opening and passing task.

We have developed a teleoperation system for a transportation task by a tracked robot with two manipulators. The Robot Operating System 2 (ROS2) was installed in the system. We newly made a new ROS2 package so that the robot can control the motors for joints of manipulators and tracks driving by a microcomputer. The operator can control a motion of the robot remotely by a joystick operation in observing images sent from a camera on the robot. The experimental results confirmed that the robot was able to perform an object-carrying task by two arms smoothly by a remote operation using the system.

We aim to develop a new gravity compensation mechanism using a counterweight system. As the load to be compensated changes, the counterweight moves to the equilibrium position to maintain balance. In addition, the center of gravity of the device equipped with the proposed mechanism can be kept within a very narrow range, thus reducing the risk of tipping over. In this paper, we fabricated a prototype of the proposed mechanism, verified the principle of the mechanism from the prototype, and examined the possibility of mounting the mechanism on the main body.

The mainstream RTK-GNSS receivers were connected to a PC via a USB COM interface, and received correction information via a PC connected to the Internet. In recent years, however, devices such as u-blox' ZED-F9P are becoming more popular, which provide a stable fix solution within the receiver module. On the other hand, the dimensions and shape of autonomous mobile robots vary from application to application, and the required position measurement accuracy also varies. In this study, we aim to build a small device that can provide the minimum RTK-GNSS receiver function by itself, and combine several of them according to the application to improve the accuracy and update speed for a scalable positioning system.

In recent years, the importance of autonomous mobile robots has been increasing in disaster situations．In this study，we developed a system that can measure the straight-line distance to obstacles in real time using a monocular camera and a line laser，and verified the effectiveness of the system using a crawler robot equipped with the system．The effectiveness of the proposed method is demonstrated by making a crawler robot equipped with a line laser and a camera move straight toward an obstacle, and comparing the distance measured from the image processing results in a dynamic environment with the actual distance between the obstacle and a crawler robot.

Most of the public infrastructure equipment currently in use was built during the period of high economic miracle and is rapidly aging. On the other hand, the number of workers who inspect those infrastructure facilities has been decreasing in recent years. In particular, the inspection work of underground buried infrastructure requires a great deal of labor and time, so the introduction of autonomous mobile robots is required from the viewpoint of improving the efficiency of inspection work and saving labor. High self-position estimation accuracy is required to identify the position by comparing the coordinates of autonomous movement and underground buried objects with the information of the ground penetrating radar. In this study, we constructed a two-dimensional relative positioning accuracy evaluation experimental system assuming inspection of underground buried objects by an autonomous mobile robot, and conducted self-position estimation experiments using an axle encoder and self-position estimation experiments using sensor fusion.

Bats realize echolocation by simple ultrasonic sensing, which is composed of one transmitter and two receivers. We are studying the various echolocation behaviors of bats by a constructive approach using autonomous vehicles equipped with broadband ultrasonic transmitters (thermophone) capable of emitting ultrasonic FM pulses that mimic bats and two receivers. Our previous behavioral studies suggested that bats flying in groups slightly change the terminal frequency (TF) of the FM pulses in order to avoid acoustic jamming caused by other individuals’ sounds.

In this study, we implemented this frequency adjustment behavior of bats as a jamming avoidance model in autonomous vehicles and conducted obstacle avoidance experiments with and without the jamming avoidance model in the acoustic interference environment. The results showed a significant improvement in collision avoidance performance, suggesting that the frequency adjustment behavior of bats could be a useful solution for practical acoustic jamming problems.

In order to reduce the burden on the elderly in heavy snowfall areas, it is important to support snow removal work with robots. In this study, we propose a motion control method to follow snow-covered road for autonomous mobile robot equipped with an RGB camera. To this end, we improve the road detection accuracy of semantic segmentation by using generative adversarial network (GAN) in snowy environment. In addition, the central point of the detected road area was extracted as sub-goal for the motion control of the mobile robot. The experimental results show that the proposed method has improved the accuracy of semantic segmentation and the mobile robot is able to follow snow-covered road stably.

The goal of this study is to develop an ultrasonic sensor system using MUSIC method which consists of an ultrasonic phased array transmitter and a microphone array receiver that we proposed, which can be mounted on a small mobile robot. Therefore, in this study, we aim to build a sensor system using the Raspberry Pi 4B, the algorithm for the MUSIC method is implemented using the Julia language, which is characterized by its fast execution speed. In this paper, we evaluate the processing time of the MUSIC method and confirm the accuracy of object position estimation on the faster processing system developed. As a result, we confirmed that it is possible to estimate the object position with high resolution and fast processing time.

Mobile robots, such as self-driving wheelchairs, need to be able to recognize bumps in door sill and gutters in order to move safely. If the robot does not recognize the bumps and moves forward at a certain speed, it may not be able to get over the bumps or may fall down. In this research, we propose a method for recognizing obstacles such as bumps and reflecting them on a local map to ensure safe autonomous movement.

This paper describes the development of an autonomous mobile robot with passive measuring wheels to provide public services in urban areas. The robot has two differential wheels and a passive wheel placed at the center of the differential wheels. The robot can estimate its correct self-position if the contact surface of the measuring wheel does not slip even if the two differential wheels slip. The slippage is detected by measuring the angular velocity of the measuring wheel, which can be used for route planning in the future work.

In this paper, we propose a method to predict terrain traversability cost using self-supervised learning for mobile robot in uneven terrain. The robot learns to predict risk of traversing the terrain by using the vibration information from the acceleration sensor as the prediction target and the color and depth image from the camera as the input. We conducted an experiment to get the datasets in a real environment and predict terrain traversability cost from terrain image at constant speed. The prediction results showed that low and medium cost could be predicted. In addition, we propose safer prediction method considering speed change.

This research aims to discuss how to obtain the number of commands needed to control a mind drone from Electroencephalogram (EEG) signals. The brain-computer interface (BCI) analyzes brain waves to identify human emotions and brain perception and control locations. To use this technology, when a specific thought or stimulus is received, its brain waves can be used as a control command for the robot, but consumer BCIs can only handle fewer commands. On the other hand, to control a drone freely requires more than seven commands. This study investigates and compares what kind of stimulus-based command classification is necessary and what kind of machine learning algorithm can obtain more than seven commands.

Recently, Robots are used in various fields. Autonomous mobile robots are one of robots that are active in society. Theses robot needs to charge with dedicated docking stations. This charging method have disadvantages in that robot’s range of motion is limited, in that putting stations take cost and so on. Therefore, we propose a new charging method, “Plugin-docking system”. But present plugin-docking has a low success rate in docking into a wall outlet. In addition, it takes a long time to finish docking. Therefore, in this research, I modified the motion of the robot to enable docking into a wall outlet, which requires less docking time.

This paper develops a visual environment voice expression system for a guide dog robot. This system enables the robot to recognize objects in its surroundings, register the scene, and generate text using CNN and LSTM methods for the guide dog robot to read. Then the robot conveys the visual scene recognition result by voice, which is generated by speech synthesis. The guide dog robot can guide the visually impaired person safely to the desired destination, as well as entertain the user by conveying various visual information through voice. The system is composed of object recognition, scene registration, text generation, and speech synthesis. The effectiveness of this system is confirmed through experiments conducted with our guide dog robot.

Recognition of the traversable area is a vital function for autonomous mobile robots. This paper proposes a probabilistic generation of a semantic occupancy grid map. The proposed method uses inverse perspective mapping (IPM) of semantic segmentation and prior occupancy grid map to recognize the traversable area. We apply the binary Bayes filter and the truncated normal distribution spatial filter to deal with the uncertainty of semantic segmentation and IPM distortion to the far side. The experiment result shows these filters improve the recognition accuracy of the traversable area.

There is a growing demand for mobile robots that can operate in inaccessible environments such as disaster sites. Such a robot that can move autonomously over unknown rough terrain needs to be able to acquire both its own position and the surrounding environment, but current robots rely mostly on external sensors. Since external sensors are susceptible to disturbances and cannot be used in some environments, a subsystem with internal sensors that are less susceptible to disturbances is necessary to operate under such circumstances. Therefore, in this research, we consider the acquisition of the surrounding environment by the internal field sensor. In this paper, we propose a method to convert the terrain information into maze information using the robot’s posture information, and to traverse the generated maze using the Adachi method. The effectiveness of the proposed method was confirmed by simulations and actual experiments.

This paper presents a force sensing system and power assist control for an omnidirectional cart with multiple active-casters. The drive wheel unit which includes differential type active-caster, 2motors, force sensing mechanisms is used to omnidirectional cart to measure the drive torque and force applied by a human operator. The force sensing system can detect force applied at any point on the cart. Based on the force information, the omnidirectional cart is controlled to assist the applied force to realize omnidirectional motion of the cart. The structure and fundamental experiments are presented on this paper.

A method searchinf for a person by focusing on the clothing of the target is proposed. YoloV3 is used to extract the area of the person the 360 degree image by RICO Theta Z1. The extracted image is compared with the photo to be searched by the cherasteristic of the person’s wear. The ResNet model is trained using the Category and Attribute Prediction Benchmark of the Deepfashion dataset, which contains images of various outfits. This model is used for discrimination by calculating the similarity between the clothing of the person in the camera and the clothing of the person in one photo.

This department is developing a crawler robot that can run autonomously in the open air. In order to prevent collisions with obstacles during autonomous driving, we have developed an obstacle detection technique using 2D-LiDAR that is robust to the vibrations and shaking characteristic of crawler robots. Our method is based on the line data of the running road surface from the irradiation data of 2D-LiDAR, and detects obstacles on the road surface from each feature. In this paper, we explain the method, clarify the problems based on the experimental results of the avoidance behavior based on the obstacle detection results, and propose solutions to the problems.

In this paper, we use an RGB-D camera to estimate and track the centroid of a person. In order for a robot to run safely in the crowd, it is important to recognize and track people with high accuracy, and the RGB-D camera is suitable for human tracking because it can not only recognize people with high accuracy based on images but also obtain distance information of the recognized people. The proposed method improves the performance of centroid estimation and tracking by removing noise from the distance information and considering the motion model of the person. The effectiveness of the proposed method is demonstrated by comparing the estimated position and labeling of a person with the true values in an experiment of simultaneous tracking of multiple persons.

Autonomous mobile robots are required to operate in a wider area of environments. When having a security robot monitor people in the environment, we thought that the robot could be operated more efficiently if it could understand the movement paths of people in the environment. In this paper, we propose a new method for obtaining movement paths information over a wide area. The presence/absence of people in the observable area, the walking speed, and the elapsed time outside the observable area are integrated to estimate the movement paths of people. We can obtain global information of movement paths from local observation information. Through experiments with security robots, we show that it is possible to improve the quality of security by obtaining trends related to human movement in the environment from estimated path information and using these trends in the robot’s patrolling strategy.

In the Lidar SLAM method, Lidar and odometory are used to generate a map of the environment around the robot and to estimate self-position probabilistically. In SLAM, it is common to use multiple sensordatas for Lidar and odometory, but when generating a map using different sensor data, i.e., sensor fusion, errors occur due to the time difference between sending and receiving sensor data. In this research, we attempt to reduce this error by performing scan matching that takes into account the arrival time delay.]

This paper describes our autonomous navigation system based on A1, a commercially available quadruped robot. We mounted a 3D Lidar and an ultra-mobile PC on the robot and ported our recognition and planning software. We used the controller provided by the manufacturer for walking control, which can be operated by giving two-dimensional abstract velocity commands. We conducted navigation experiments by participating in Tsukuba Challenge 2021, and we encountered a number of failure cases, including sudden falls caused by overheating. We overcame those problems, and our robot eventually completed a 2 km autonomous walk in a real urban environment, except for manual battery exchanges.

Semantics such as object classes is typically denoted by a discrete variable. To probabilistically leverage semantics in localization, the discrete variable must be introduced in the likelihood calculation considering continuous geometric relation without loss of probabilistic manner. This paper proposes a novel localization method that enables to probabilistically handle semantics. The proposed method uses Dirichlet distribution in the likelihood calculation. The hyperparameters of Dirichlet distribution are determined using geometric relation. As a result, semantics can be probabilistically handled and the likelihood can be calculated while coping with uncertainty of object recognition. Experimental results show that semantics can be utilized for improving localization accuracy while preventing performance reduction even when object recognition has been failed.

Autonomous mobile robots need generated operation planning to control in advance. Several robot vendors have been developing operation control system which can generate operation planning to control robots. If the interface is different between robots and system, the development efficiency drops. Toshiba formulated a common interface; AMR-IF (autonomous mobile robot interface) and developed AMR-IF-UI which implements this interface. This system generates operation planning to input some commands in the form of a list. However, operators are required technical knowledge and labors. Therefore, we suggest a supporting software that operators can generate operation planning to connect to each function block by arrows instead of command inputs.

For automatic driving, a three-dimensional point cloud with highly accurate information around the road is necessary, and SLAM technology has been attracting attention. Therefore, in this study, we focused on the merging of three-dimensional point cloud maps, and aimed to construct a wide-area three-dimensional point cloud map by merging three-dimensional point cloud maps. However, in order to join the point cloud maps properly, it is necessary to solve the trade-off between continuity and positional accuracy. Therefore, we proposed a method to solve the trade-off problem by optimizing the range of absolute position constraints. As a result of evaluation tests, it was confirmed that both continuity and positional accuracy can be maintained by adjusting the range of absolute position constraint, and the effectiveness of the proposed method was confirmed.

In recent years, automatic driving has been attracting attention, and various companies and institutions have been actively conducting research. In particular, Localization technology using 3D point cloud map and LiDAR (Light Detection and Ranging) has attracted much attention. However, it is known that localization using 3D point cloud map and LiDAR may cause misconvergence and increase of localization error when the terrain has no feature or similarity.

Therefore, it is important to guarantee the accuracy of localizations and 3D point cloud maps. In this study, we propose a new method to analyze the error tendency of localization using 3D point clouds from the viewpoint of the reliability of matching, and analyze the error characteristics of localization under typical environments where localization is often failed. The purpose of this paper is to analyze the error characteristics of localization in typical environments where localization tends to fail.

In autonomous vehicles, it is desirable to drive while estimating the error state in places where the error in the localization using LiDAR becomes large. In this study, we have developed a new method to estimate the error state of the localization in real time, which enables more accurate localization in real environments. The effectiveness of the proposed method was verified in places including environments with few features and where there are errors in the measurement of 3D point clouds. It was confirmed that the enlargement of the pose error in those places could be grasped by the error ellipse calculated by the proposed method.

In recent years, there has been a lot of technological development toward the realization of fully automated driving. In fully automated driving, awareness of the surrounding environment is essential, and vehicles must be equipped with sensors and cameras for this purpose, as well as receivers to acquire positional information. In this study, we considered the application of these techniques to the observation of snow cover. At present, the main method of observing snow cover is by using snow depth gauges installed at each observatory. However, this observation method does not allow us to determine the amount of snow cover at locations far from the snow depth gauge. Therefore, in order to obtain the road surface conditions under snowy conditions, we will attempt to create a three-dimensional map under snowy conditions using a mobile robot equipped with a laser range finder, RTK-GNSS (Real Time Kinematic-GNSS), and IMU (Inertial Measurement Unit).

When RTK-GNSS is used for tall large vehicles, there is a problem that the estimated vehicle position is easily affected by the attitude angle. In addition, the acceleration error in the IMU used for attitude angle estimation makes it difficult to estimate the attitude angle with high accuracy. In this paper, we focus on the roll angle and pitch angle, and propose a method to estimate the attitude angle accurately and robustly by using LiDAR and IMU to compensate for the IMU error. In addition, an evaluation test of the proposed method is conducted to confirm its effectiveness in attitude angle estimation.