Currently in the agricultural machinery field, various types of robot tractor and automatic work machine are being developed. Among them, the automatic driving mechanism using the GPS-RTK is rapidly spreading. However, due to the use of radio waves, radio interference has occurred which directly leads to the accuracy of automatic operation. In this research, we focused on the experimental research for developing a system that supplements the conventional automatic steering system. In the outline of the system, ridges are recognized by image processing, and a route is generated based on this. As a result, accuracy is not high, but it was proved that traveling along the ridges in the fields is possible with information from the image.

In this research, we developed a multi-purpose traveling vehicle for paddy fields with the aim of fostering agricultural engineering human resources and evaluate its traveling performance. The vehicle is a two-wheeled vehicle capable of traveling paddy fields in the flooded period and has a configuration that makes it easy to implement new functions by replacing float parts. The prototype was used in the PBL in order to consider the effective usage. As the result, a consideration to use flexibly configurable robot from the viewpoint of the user is effective for fostering robot user in agriculture was given.

This paper describes methods of recognizing cucumbers for autonomous cropping of their fruits. Cucumber recognition from color images by conventional methods is challenging because their leaves and stems have similar colors to that of fruits to be detected. Therefore, we adopted deep learning methods to recognize them. In particular, we compared two types of deep learning architecture. One is U-Net, a network for semantic segmentation, and the other is YOLO, a network for object detection. As a result of the comparison based on precision and recall, we found that U-Net has a comparable performance to YOLO in terms of cucumber fruit detection.

The purpose of this study is to develop a remote controlled weeding robot using a portable terminal. The developed weeding robots are a wheel type weeding robot and two spherical type weeding robots. The wheel type weeding robot has four-wheel drive controlling device which improves the traveling stability for paddy field. The spherical type weeding robots roll on the surface of water. In this paper, we carried out the experiments using Wi-Fi or Bluetooth communication method from the portable terminal. The effectiveness of the developed weeding robots are verified by the experiments.

In this paper, we propose a robust algorithm to detect landmarks in a vineyard field. The algorithm only uses data from laser range sensor to accurately detect pillars supporting the grape plants in the presence of noise. Robust detection of these landmarks is important for the autonomous task execution by a robot in vineyards. The detected pillars are also used for accurate localization and mapping. Moreover, based on accurate pillar detection, semantic data logging in vineyard becomes feasible. Experiments in both simulation and actual vineyard fields are presented. Results show that the proposed algorithm can accurately detect features with inexpensive Lidar sensors to enable accurate robot operation in vineyards.

We describe a harvesting robot that grabs fruits and plucks them. To pluck the fruits, it is necessary to reproduce the skillful human hand motion of pulling while rotating. Furthermore, plucking motions vary greatly depending on conditions of the fruits and branches. We focus on the force from the branches and determine the conditions based on the ease of bending of the branch. In this study, the ease of bending is defined as the force from the branch per unit length when the gripped fruit is pulled down slightly. Moreover, using probabilistic learning and combining of plucking motions taught directly by a human, new motions are generated according to the ease of bending of the branch. Thus, plucking motions similar to that of a human hand can be generated according to the situation with fewer teachings. The effectiveness of the proposed method is demonstrated by experiments.

In recent years, precision agriculture has gained a lot of momentum in automated agriculture research. With increasing aging population around the world and shortage of labor for agricultural tasks, automated farming is believed to play a key role to sustain the agricultural growth. In this paper, we present an autonomous robot navigation for vineyard monitoring and maintenance. To mark the position of tree trunks and rows, mapping of the vineyard using range sensor is presented. The rows are identified for navigating the robot and path planning is devised to be used for monitoring the field. The experiments were performed in actual scenarios, and the results are presented.

In tree branch structure recognition for fruits identification and branch pruning, it is necessary to compensate missing 3-dimensional data. In this paper, we propose an algorithm to detect 3-dimensional information shortage and an algorithm to determine the target position and posture that is expected to compensate the shortage. We propose a flight planning method of a drone that flies toward the target position and posture for compensating missing 3-dimensional data. We verified each algorithm and carried out the flight simulation of the drone. We also compared tree branch structure recognition results and confirmed that missing information was compensated.

This manuscript describes an assist robot to harvest citrus fruits. The purpose of the robot is to carry citrus fruits between the fruit trees and a monorail automatically. This monorail carries a container which is filled by the fruits to the base of mountain. In order to carry the fruits without touching the branches of trees, it is desirable to make the size of robot small. On the other hands, it is requested to move the fruits to a container on a monorail. The height of the container is about 1m. Therefore, the robot needs to carry and lift up the fruits. This robot has the pantograph mechanism with a linear moving. actuator. This mechanism can lift up the weight of 15 kg stably. In this paper, the performance of the mechanism is shown.

A tomato harvesting robot that can change the hardware configuration depending on the greenhouse has few development cases, and such robots are also important in terms of operation and development cost. We focus on the tomato harvesting robot that can change only the hardware configuration without changing the basically software configuration. In developing such a robot, it is important to modularize the hardware as the component. Building a distributed system makes it unnecessary to change the software configuration even if the hardware and sensors are changed, and it is possible to develop different robots only by changing the modules. In this study, three tomato harvesting robots with different hardware configurations are developed based on the above development concept. Two types of manipulators and moving mechanisms are introduced. We also report harvest experiments using two manipulators.

By utilizing the Fin Ray® effect, we develop soft robotic hands that can grasp leaf vegetables such as spinach of hydroponic culture without damage only by position control. As a preliminary stage of the development, we fabricated two types of the hands using a silicon material. The gripping force and pull-out force were investigated through measurement experiments. As an experimental result of grasping of spinach, the soft hands realized grasping motion without damaging of spinach.

In recent years, drastic decreases of agricultural workers and their aging become a significant problem. Application of the robotic technology into agricultural fields is being required for labor-saving. In particular, harvesting operations for spinach almost depend on manual works. In this research, we aim to develop a harvesting robot for hydroponic cultivation plant, where cultivation environment such as pallets and benches are standardized. In this paper, we conducted preliminary experiments for considering the specification of the robotic hands. Based on the specification, the prototype was designed and fabricated. The grasping motion of spinach was investigated through experiments. And we conducted a harvest experiment on the farm.

This paper describes the development of a harvest time prediction system based on color information of tomato. For stable supply of agricultural crops, it is important that prediction of harvest time and amount of harvest. Generally, the harvest time of agricultural crops has various factors such as air temperature, soil temperature, humidity, and so on. It is difficult to measure all of these factors for each crops. In this research, we proposed a harvest time prediction system based on Long Short-Term Memory (LSTM). The LSTM predicts the time series data of tomato's color information. Regarding the accuracy of color prediction and how many input data is needed, we comparing the single layered LSTM with double layered LSTM in this paper. The double layered LSTM achieved the more accuracy prediction with less input data than single layered LSTM.

In this research, we developed a vertically articulated manipulator and a vision system for tomato harvesting. The developed manipulator and vision system were evaluated in tomato robot competition, the robot achieved the tomato harvesting by remote operation and automatically harvesting motion. Regarding the decision making of harvesting, proposed vision algorithm shows better performance for multiple tomatoes.

In fruit harvesting farmhouses where slopes are steep like mountainous areas, it is difficult to manage the trees and there is a problem of leaving fruits of high places. In the conventional robot hand, when it is out of the center of the hand, it cannot exert a gripping force and it gives a local force to the harvested product and hurts the fruit. Therefore, in this study, we introduce a mechanism that can grasp even if the shape and size of the fruit change, we can harvest fruits at a high position in an unstable place such as slope and develop a portable harvesting robot hand.

In recent years, automated driving and AGVs are actively researched and developed. The important technique is Simultaneous Localization and Mapping (SLAM) which creates maps and self-location estimation at simultaneously. There are several kinds of SLAM depending on the sensor and algorithm used, ORB-SLAM uses only monocular camera. Since it is cheaper than other sensors, it is easy to board it into AGV etc. However, the calculation amount is large. Although high performance PCs is required, it is not able to board it into AGV etc., so we focused on FPGA. The FPGA development environment is also actively developed, and the development with high-level synthesis has become realistic. Therefore, in this research, in order to improve FPGA development efficiency, we implemented ORB-SLAM from open source software to FPGA.

In this paper, we propose a route planning method to keep the accurate localization for unmanned aerial vehicles (UAV) which use a camera and an inertial measurement unit (IMU). The dead reckoning with the IMU and visual odometry (VO) using the camera can be used for the self-localization in an unknown and GPS-denied environment. Both techniques cause estimation errors with motion, but the VO has less error accumulation than the dead reckoning. Therefore, we developed a method of modeling the predicted routes to calculate the error variance as an objective function. In addition, this approach enables determination of the posture of a UAV equipped with a camera with a high number of degrees of freedom. The simulation results show the accuracy of the proposed method, and confirm that it has a certain error improvement effect in any environment.

An environment map is essential information for autonomous mobile robots. In this map, a robot plans the path, localizes the position, and controls the motion. In general, the map for robots is generated through a so-called SLAM technique, simultaneous localization and mapping. However, it is difficult to generate the entire environment map. For this issue, we focused on a geographic information system, GIS, map. Furthermore, we propose a localization method for a robot in the GIS map. This method is composed of two localization phases: global localization using GPS and precise localization using 3D LiDAR. In the experiments, we discuss the effectiveness of the proposed method in consideration of the localization and navigations results.

Social advancement of robots has become remarkable, and there are many opportunities to see in daily life. Therefore, in this paper we proposed a new method for creating an environmental map of a robot. In the conventional method, object information was included after map creation in order to include object information. The method proposed in this paper automates the creation of environment map including object information by creating deep learning in real time when creating the environment map. It is thought that this makes it possible to give general versatility to robot action decisions. Experimental results showed the effectiveness of the proposed method.

An autonomous mobile robot in a dynamic environment needs to avoid static obstacles and also needs to perform path planning while considering dynamic obstacles. In this paper, a path planning method is proposed for avoiding encounters with moving objects by incorporating an abundance map that represents the possibility of existence of previously observed moving objects. The proposed method can generate a path that avoids areas with high human traffic, which reduces the probability of obstacle encounters and path re-planning. Results of the evaluation experiment indicate that compared to conventional method, the proposed method can generate a shorter path in a dynamic environment having high human traffic.

The goal of our research is to apply the Mecanum Wheel robot as an automatic conveying robot in a narrow space. The Mechanum Wheel is a wheel covered with a barrel whose surface is tilted by 45 degrees with respect to the axle, allowing it to move diagonally in addition to straight ahead. By attaching Mecanum Wheels in the same way as for ordinary vehicles, Mecanum Wheel robot can move omnidirectional by combination of four Mecanum Wheels.

However, Mecanum Wheel has a problem that the amount of movement changes due to the influence of the friction with the floor surface. In this paper, we describe the influence of frictional force on the Mecanum Wheel.

In recent years, researches on autonomous mobile robots that work in unknown environments such as exploration robots and rescue robots are actively conducted. In previous studies, external sensors were often used to recognize external conditions, however, they are often unreliable in sandy or snowy environments. The purpose of this study is to estimate the displacement of a wheel-legged robot with high accuracy by using only the information obtained from the internal sensor. In this research, we focus on using only inexpensive internal sensors such as an acceleration sensor, a gyroscope, and a torque sensor (using strain gauge for the actual machine). To investigate the performance of the proposed displacement estimation method, experiments of the robot running through several environments were conducted using ROS and Gazebo. By the simulations, it was confirmed by the proposed method that displacement estimation accuracy has been improved.

When inspecting infrastructure buried in the ground such as pipeline, it is necessary to investigate the existence of other buried objects. As a method studied for that purpose, there is a method in which a hand-operated cart with a ground penetrating radar is moved on the ground in an area to be investigated for measurement. It is expected that information can be effectively utilized by estimating the position of the cart and correlating it with the data of the underground radar. In this paper we propose self-position estimation method using small image sensor and acceleration sensor. Furthermore, we verify the effectiveness of the proposed method by demonstration experiment.

This paper describes the new ideas for improving methods of the estimation accuracy of localization that uses image sensors and QR codes for videos. The ideas the authors describe localization methods which take accounts of continuity of frames in the video and use multiple QR codes filmed in the image sensor ’s view.

As one part of the decommission tasks of Fukushima nuclear power plant, efficient and appropriate operation of removal of fuel debris requires essential information of distribution and characteristics of fuel debris. A robotic remote measurement system using an ultrasonic sensor called RhinoUS was developed to detect the distribution of the fuel debris and localize the unknown water leakages in the primary containment vessels (PCV) of the No.1 reactor. Experiments are carried out to evaluate the localization accuracy on the robot pose of the visual odometry method.

The results turn out that we could acquire characteristics of the fuel debris and achieved localization error less than 15mm/1m in complex motion, and generate smooth grating map in bird view. We conclude that the visual odometry localization method might serve as a feasible solution to assist the robot localization on grating.

In most cases, many sensors are used in autonomous robots for self-position estimation to acquire detailed coordinate values, create a dense map, and implement map-matching. However, though human being doesn’t use coordinate values, it is easy to head toward the destination. There are many parameters human being uses for self-position estimation. Among them, we focus on scenery change in this paper, and propose a method of self-position estimation using sparse scenery information. In order to detect scenery change, we quantitated scenery information by using SegNet and named the value VQ (Visual Quality). First, we verified that the similarity of a VQ wave would be high in the same rout. Next, we used VQ and classes of semantic segmentation, and SIFT (Scale-Invariant Feature Transform) to implement selfposition estimation. As a result, the accuracy was enough and VQ can be used as a parameter for self-position estimation.

A planetary exploration rover is required to run on the designated route in an environment where slip and stack easily occurs. Therefore, self-location estimation is important for the planetary exploration rovers, but the current estimation methods have disadvantages and it is necessary to improve the accuracy of self-location estimation. In this research, we propose a method to estimate the slip ratio by detecting the rut which is the running trace of the rover, and verify its effectiveness. As a result, the slip ratio estimation method using rut was effective, but improvement of rut detection method has been necessary.

In recent years, automation of mobile robots that can work with people is required. It is important that the mobile robots move based on the map created first rather than moving while creating the map, because robots need to move according to the administrator’s instructions. For that reason, it is necessary to be able to move even if the surrounding environment changes, such as changes of illuminance and movement of objects, since the map was first created. We developed a self-localization estimation algorithm robust to environmental changes using monocular camera suitable for recognition. We also implemented this algorithm on commercial conveying cart and confirmed automatic moving, so we report it.

Mobility expected to autonomous in recent years. However, the current navigation is a problem that is unnatural. In the proposed method, while localizing, driving data accumulated by people is collected and clustering. The feature of the proposed method is that it can register human senses on the map. We aim to adapt to the sensitivity of the individual by conducting navigation for each category. As an evaluation test, we used modified electric wheelchair. We collected data 10 people indoors. Even indoors, we can distinguish driving tendencies and confirmed the effectiveness of the proposed method. As a schedule for the future, we will establish classification methods and conduct evaluation tests in outdoor environments.

Recently, autonomous mobile robots become popular as a response to labor shortage especially in airport and large commercial complexes. However, it is difficult to recognize global self-position by previous localization method in indoor environments. Also, the previous method is difficult to introduce because it makes the design of the robot bad and to introduce new equipment in the facility. This paper presents the global self-position recognition method which does not deteriorate the robot design and does not require a new facility.

This paper proposed a method of improvement of Localization Accuracy of Cloud-based Localization System using 3D CAD map (Universal Map) by gradient dilation matching. This system only prepares line segment datebase images made from 3D wireframe map in advance to estimate location. This system estimates a location by matching sensing image and database images. But, if this system uses conventional similarity operation, there was a case in which image matching failure occurs. So, I developed gradient dilation image which is converted from line segment images and new similarity operation to reduce matching failure. gradient dilation image is dilated line segments and give a gradient to density. gradient dilation matching improved localization accuracy by 90% in Universal Map.

In this paper, we propose a solution method of kidnapped problem. In the case where the tracking of the self position becomes impossible after losing the feature point, we will develop a method to identify the position and restore the route by using a cloud type system that can find global localization.

For autonomous mobile robots, self-localization is an important problem. In order to reduce the calculation cost for self-localization, we propose a self-localization method using the reliability map of sensory information in known environment. In past studies, we evaluated our proposed methods by the locatization experiment with a real-robot in a real environment, however, the verification was insufficient. In this paper, we evaluate our proposed methods by the localization experiment with a real-robot in various real environments.

In this paper, a robust mapping method considering the space status of real-life environment is proposed. A 2D map is generated by Simultaneous Localization and Mapping (SLAM) based on gmapping. The objects in the 3D space are recognized by deep learning method YOLO and reflected to the 2D map. The 2D map contains the 3D shape information of the objects so that the robot can plan its path more accurately without collisions with objects and entering the places where the spaces are potentially occupied by the objects (e.g. the spaces under desks).

In controlling the mobile robots, self-localization is an important elemental technology. As a self-localization method of the wheel type mobile robots, there is a dead reckoning method which estimates a movement distance from a rotation amount calculated based on values from an encoder mounted on wheels or motors. However, in practice,there is an error caused by external factors such as idling of the wheels. As one of choices to reduce the error, methods of estimating the self position by using relative positions of other robots are introduced, especially in swarm robot environment. Although these methods can effectively utilize the existence of surrounding robots, they are often limited to verification in simulation. One of the reasons is the difficulty in developing communication devices for swarm robot environments. The spatially seamless local communication system we have developed is effective as a device for that purpose. However, it is difficult to mount it on small robots due to its large size and its complicated structure. In this paper, we propose the communication system redesigned to from a structure that is compact and suitable for self-localization.

To prevent occurring robot kidnapped problem, it is important to evaluate likelihood of each particle considering an environmental change. Moving objects are one of the main causes of environmental change and each objects has each own movability. For example, chair has high movability because it is designed to move and interacts often with humans. On the other hand, walls or shelves have low movability because they are designed not to move and interact less often with humans. So, in this study, we define classes of objects and movability of them. We propose a localization approach that focuses on the association between sensor information obtained from objects whose movability is low and prior map by considering classes and movability in order to prevent occurring robot kidnapped problem.

The purpose of this research is the sharing of mapping data on multiple units for environment mapping by the SLAM method and improvement of mapping accuracy. There are many obstacles, which are dynamic or static, in a living space. In addition, the robot needs to decide the information from the external sensor attached to the robot. When a small robot works in a room, creating a map is inefficient. Thus, we thought about mapping and sharing data with multiple robots. We tried to make robots created maps efficiently by sharing data. We experimented on environment mapping using a small sensor of a small mobile robot and examined accuracy of the localization. As a result, we obtained accuracy of robot's self-position estimation and accuracy of a small distance measuring sensor used for environmental map construction experiment.

Lifting and carrying heavy loads often cause serious low back pain especially at work sites of the nursing, the civil engineering and the logistics. To alleviate and even solve this problem we have developed a backdrivable waist assisting robot. In this paper, we described the control system for the waist assisting robot. Further, we evaluated the performance of the control by measuring it’s output torque. The experimental results verified the effectiveness of the developed waist assistive robot. Meanwhile, we analyzed the causes that resulted in low efficiency of the power transmission system of the robot.

Recently, in Japan, a burden for labor is increasing due to the aging society. To solve these problem, assistive suits are proposed for one of the solutions. In the previous study, an assistive suit, Airsist I has been developed that is based on the variable viscoelasticity. However, the motion judgment interface for Airsist I has not been studied. In this study, joint angle and joint angular velocity are used to judge the motion of the wearer, because Airsist I has a high backdrivability. This paper reports the motion judgment algorithm that is based on the wearer’s joint angle, and joint angular velocity, and experiments are conducted to evaluate the utility of the algorithm.

In this paper, an assistance method that controls the joint stiffness and antagonized angle using variable elastic elements is proposed. The proposed system changes the joint stiffness and antagonized angle so that they correspond to the phase of movement, and performs movement assistance in cooperation with the wearer. As a structurally variable configuration of stiffness and antagonized angle, a joint structure in which an artificial muscle and tension spring are antagonistically arranged is proposed. A mathematical model of one side spring antagonized joint is proposed and its characteristics is examined.

Low back pain is an important international health problem. Factors of back pain have a burden on the lumbar region due to the handling of heavy objects. The Japanese Ministry of Health, Labor and Welfare recommends Squat Lifting as a method with less load on the waist. However, Squat Lifting is not used much because it has a large workload. Therefore, we aimed to develop a power assist suit for spreading Squat Lifting. In this paper, we first describe a device with a knee joint extension mechanism by a leaf spring and its problems in a previous study. Next, we propose a sliding mechanism as a solution. Then, a three-dimensional motion analysis for determining the necessary assist torque during Squat Lifting will be described. Finally, we will explain the leaf spring characteristic test which determines the specification of leaf spring.

Many institutes and companies have been developed wearable robots for people with disability. However most of wearable robots require a walker or canes for maintenance of balance and lateral weight bearing. Consequently, the user gets large physical burden. In this paper, we have developed a new wearable robot with a weight bearing mechanism. Our wearable robot assists not only the hip, knee and ankle joint movement but also the lateral movement of the pelvis by means of a parallel link mechanism. Carrying out experiments of our wearable robot with normal subject, we confirmed that the lateral weight bearing could be performed by our mechanism. We also confirmed that our wearable robot was able to assist walking.

In this research, we report the evaluation of soft power support suit for upper limbs using ‘18 Weave’muscles. We have developed a novel active weave named‘18 Weave’using a thin McKibben muscle and orthoses based on a body measurement. We have protyped asssist suits based on these ideas. In this paper, we evaluate the power suit by electromyograph. Measurements of EMG confirmed significant decreases in iEMGs of the anterior deltoid, posterior deltoid, biceps brachii, and triceps brachii muscles.

The purpose of this study is to develop a lighter and smaller auxiliary device that can assist the user’s waist to reduce burden a half-sitting posture of a worker. The auxiliary device consists of an upper half body part and a lower half body part. In this paper, we used a ratchet mechanism and a power spring in order to use a wire. The ratchet mechanism is used to adjust the half-sitting posture. The effectiveness of the proposed support device is verified by the result of the experiments.

Recently, powered exoskeletons for strengthening have been used in heavy work like construction industry. However, most of the powered exoskeletons can only strengthen a part of the body of a user. A small and light actuator which generates great force is required for the powered exoskeletons strengthening the whole body. It is expected that a thin hydraulic artificial muscle is such a high efficient actuator. In this paper, we developed the thin hydraulic artificial muscle that has a mass of 22 g/m and a diameter of 6 mm. It has 1262 N generation force in applying 5 MPa. We also developed and evaluated a powered exoskeleton for strengthening hands using the thin hydraulic artificial muscle. The result showed that the artificial muscle strengthened the grip force of a finger about 300 N under a pressure of 3 MPa. Therefore the thin hydraulic artificial muscle is effective for strengthening powered exoskeletons.

In this study, prototype of an electric assist suit which can be used in both walking assist for elderly people and lifting motion in transfer work for caregiver was developed, and especially the walking assist for elderly people is discussed. In order to evaluate the developed assist suit in the walking assist for elderly people, young subjects wearing a kit to experience elderly people were regarded as imitational elderly people. To demonstrate this validity, the evaluation experiment of the elderly experience kit was conducted, and usefulness of the kit was verified through experimental result. Then, the verification experiment of the developed assist suit on the walking assist was conducted to the young subjects wearing the elderly experience kit. From the experimental result, increase in swing up angle of thigh was observed. Thus, effectiveness of the walking assist function in the developed assist suit was verified.

Physical assistant robots are developed for daily living activities. However, the risk of fall is a major concern of physical assistant robots. In this study, we focus on curving motion, because fall is likely to occur during turning owing due to restricted degrees of freedom of joints. We developed new physical assistant robot, which could adjust the degree of freedom, and compared the curving motion. As a result, some subjects had a decrease in step length, when the hip rotation was restricted while curving. Additionally, the analysis of gait stability suggested that restriction affected the stability when they were curving.

Power assist devices are attracting remarkable attention as they can reduce the burden of workers engaged in heavy labor: nursing care, manufacturing, logistics, and farming. However, the effect of power assist devices in reducing labor’s fatigue has not been sufficiently evaluated. The purpose of this study is to quantitatively evaluate the assistive effect of power assist suit in terms of energy consumption. In the experiment, we particularly focused on the effect of Muscle suit® which was developed to reduce the burden on back in lifting heavy weight. The measurement results of energy consumption show that Muscle suit can reduce labor’s fatigue from the perspective of energy metabolism.

L-shaped grippers typically used for power assist suits have a problem in stably holding objects of arbitrary shape. To solve this problem, we developed a gripper with a ratchet mechanism consisting of linear motion and rotational motion parts. The gripper can be deformed according to the shape of the wearer’s hand and the shape of the object to be held, and the shape of the gripper can be fixed due to the rachet mechanism, thus objects of arbitrary shape can be held stably by this gripper and the burden of the wearers is reduced. We carried out experiments where four types of objects were held by the developed gripper. We showed that the developed gripper has reduced the wearer’s burden for two types of objects among four.

This paper describes the design of the legs of a small biped humanoid robot. The purpose of this study is to develop a biped humanoid robot system with a more efficient walking motion. The legs of the biped robot are constructed to be compact and lightweight to facilitate adaption to a human environment. It is comparable to the leg length of a Japanese average 10-year old child. The generation of a walking pattern to walk on a flat floor is also discussed. Through a walking experiment on a level floor, the effectiveness of the mechanism of the legs is confirmed.