In order to reduce the weight of industrial robots，collaborative robots and improve force control performance, the tip of the motor output shaft and reducer output shaft may be configured with low-rigidity mechanical parts. However, there is a problem that the mechanical resonance frequency decreases and vibration is likely to occur during high-speed operation. To solve this problem, a method of suppressing vibration by feeding back the torsional torque of low-rigidity mechanical parts is known, but self-excited vibration and steady deviation due to estimation error may occur. Therefore, in this paper, we proposed a torsion torque estimator that uses both the motor angle and the load angle via low-rigidity machine parts and has a backlash model. The effectiveness of the proposed method was shown in simulations and experiments of load-side angle control of a motor with a reducer.

In the conventional, Simultaneous Localization and Mapping (SLAM), it is difficult to calculate the camera movement using the image features in the environment with less features. To cope with this problem, we address this research proposes a method coordinating SLAM and AR marker recognition in a featureless space. We assume a processing/inspection work of a large structure by a mobile manipulator. We use a depth sensor that executes 3D-SLAM in the horizontal direction and an RBG camera that recognizes AR markers in the overhead direction, and show real-time detection of errors that occur during SLAM execution using AR markers.

In this study, we developed a system that detects a specified parts box from a shelf and pulls out the parts box using YOLACT, an object detection method. We found that the proposed method can detect various shapes and colors of component boxes and pull them out using a robot arm. We also found that the proposed method can detect the grasping position and pull out the parts regardless of the presence or absence of unlearned part boxes or parts. However, in the case of a colorless transparent parts box, the depth information was not sufficient, so the parts moved to a different position from the detected grasping position, and the success rate of the withdrawal operation decreased.

Aircraft parts have different shapes depending on the part to be assembled, and a wide variety of parts are handled. At the aircraft production site, after the parts to be used are taken out of the parts box, they are checked by placing them in a "sugatami" (parts-kit), which has an indentation molded into the shape of the part to prevent misplacement. Currently, these processes are done manually, which leads to a shortage of manpower and a decrease in work efficiency. Therefore, there is a need to automate these processes using robots. In this study, we aimed to automate the grasping and fitting of parts by using template-matching.

Because disaster areas have various road conditions, wheel characteristics must be changed to suite to them. Since it is inefficient to change wheels at the disaster areas, a single wheel is required to have multiple characteristics. Therefore, an airless variable rigidity wheel for driving in disaster areas is proposed. It is equipped with a variable rigidity mechanism using a belt drive. In this drive system, the petal-shaped parts that necessary to change the wheel characteristics can be interlocked. In this way, the wheel can be given the characteristics of stiffening and softening. Then, a single-wheel driving experiment is conducted on soft ground with a traction load to confirm the driving performance. As a result, it is shown that the switching of the wheel characteristics has the effect of preventing the wheel from getting stuck.

From the viewpoint of disaster prevention, it is expected the unmanned automation of daily inspection/health monitoring and diagnostic work for facilities/structures in places where it is difficult to let the inspectors going such as offshore plant and dangerous places. By adopting remote controlled/autonomous robot, regarding the occurrence of abnormality due to human factors, breakage of facilities due to aging of facilities, malfunction caused by them, accidents by increasing the frequency of periodic inspection, it becomes possible to prevent it in advance. In addition, introduction of the robot makes it possible to conduct inspection work even during operation, so it is expected that the availability factor of facilities in such environment will be improved. This paper described about the plant disaster prevention challenge of the World Robot Summit 2020 disaster robotics category held at the Fukushima Robot Test Field in October 2021. The competition concept and the competition rule were introduced. The analysis of the competition results were shown and considered based on competition results. Finally, the lessons learned was mentioned.

When a severe natural disaster such as earthquakes occurs, numerous victims may be trapped under debris in vast areas. In the CURSOR project, we develop a search and rescue robot system to support first responders to make collect and rapid decisions. The system deploys many mobile robots to search inside the debris from aerial drones. However, it is challenging to move around on debris for such small robots, which can be carried and deployed by drones. Herein, we report the development of a search robot that loads mono-wheeled flexible tracks, previously proposed, which has a high capavility to overcome obstacles.

When large disasters such as earthquakes occur, a large number of victims may be trapped under collapsed buildings. It is necessary to perform safe and efficient search and rescue for the first responders, however, it is technically not easy for their actual activities. In the CURSOR project, a large number of small rescue robots with advanced sensors are transported by drones to search wide areas of the disaster site. Here we show the development of a small robot called SMURF: a soft miniaturized underground robotic finder for searching under-rubble. Basic concept and hardware design to apply to debris are illustrated. Some mobility tests were conducted to study the mobility performances.

We have proposed a virtual direct operation interface for teleoperation of mobile manipulators. In this study, we conducted an experiment for four subjects to teleoperate a robot and perform a task. As a result, when the VIVE controller is used for the proposed interface, the achievement time of the manipulation task is reduced. Although the physical demand to the operator is increased, the mental demand can be reduced in the case of the VIVE controller compared with the game pad. However, in the direct operation using the gripping sense presentation device, not only the working time but also the mental and physical demands were increased compared with the VIVE controller for all tasks. We found that it is necessary to change the input and haptic method and install new interaction functions for the proposed interface.

This paper describes the compliance control system design problem of the sub-crawler rotary joint for the ground-adaptive crawler robot. It is shown that the rotation angle of the sub-crawler greatly contributes to ground adaptability due to the relationship between the moment and propulsion force generated by the interaction between the robot and obstacles in the robot’s pushing-up sequence of the ground-adaptive crawler, and the maximum rotation angle of the sub-crawler at the moment when the robot body is lifted is determined by the compliance of the sub-crawler rotation joint. We propose a control system design method based on the relationship between the rotational stiffness of the sub-crawler rotary joint and the natural angular frequency of the compliance control system. As a result of verifying the proposed control system design method by simulation and experiment, its validity was confirmed.

Rescue robots are becoming more and more important because there are a lot of natural disasters in Japan. In order to enable rapid rescue operations, it’s necessary to introduce 4-Wheel UGV on disaster area. Since there are some problems with conventional wheels, it’s necessary to develop special wheels which can adapt rigid ground and loose ground such environments of disaster area. In this study, the Airless variable-rigidity wheels using Sickle-shaped linear mechanism is proposed. Then, the effectiveness of the wheels on driving course that combines different road surfaces was verified. From experimental results, it has been indicated that the running performance was improved by changing of the wheels’ rigidity while driving.

It has been reported that the driver workload is a cause of drowsiness, fatigue, and accidents of the driver, and a system for estimating the driver workload is required. In the conventional study, we have developed a system that estimates the current and future driver workload by using LSTM (Long Short-Term Memory). However, this system used an ambiguous subjective evaluation index for the teacher label, which had the problem of adversely affecting the discrimination accuracy. Therefore, in this study, we propose a system that can estimate the current and future driver workload with high accuracy by removing the ambiguity of subjective evaluation using semi-supervised learning and logistic regression. The experimental results show that the proposed system is superior in discrimination accuracy to the conventional study.

For object recognition using camera-captured images, coarse weather condition such as fog / rain causes edge degradation of the target object and adversely affects the detection accuracy. Aiming at constructing a “Minimal Risk Maneuver: MRM” for automated driving systems, we examined applying an algorithm based on "Dark Channel Prior" to improve the identification accuracy. Experimental results with the algorithm depicted the feasibility of the algorithm in terms of our aim. Further development with relevant techniques would be also discussed.

In order to realize safe automatic driving in urban areas where many kinds of moving objects move freely, it is necessary to predict the future behavior of moving objects. We propose a new trajectory prediction method based on the encoder-decoder framework, which can consider the interaction of moving objects in many aspects. By using labels and angular velocities for each moving object as inputs to the model and incorporating the interaction of moving objects using multi-head attention, the prediction accuracy is improved from the previous method.

In this paper, a novel electric lift cart with a constant driving force mechanism is developed in order to help labors. The proposed new lifting mechanism is able to lift loads with constant driving force regardless of the height of lift, and can lift heavy objects using a small motor. In terms of control, a sensorless force estimation method is proposed instead of a force sensor. The proposed method is based on a model of a lift mechanism, and experiments were conducted using a simplified miniature model for verification. As a result, the weights of the dumbbells are well estimated when it was placed on the platform. In the future, we will implement the method on an actual lift cart and aim to develop a method that integrates shock absorption and height control.

In the case of laying a wired communication cable by using the mobile robot, the problem is that the cable is broken due to excessive tension caused by the acceleration/deceleration of the robot. In this research, we aim to establish a drum rotation control method according to cable tension variation by a lever-type tension measurement mechanism for an auto reel device mounted on a mobile robot. This paper described the modeling method of the auto reel system, the controller design based on the concept of compliance control, and the implementation results of the designed control system. Automatic cable pay-out/winding control could be realized with the proposed method. However, as a result of confirming the control system characteristics, it came to clear that it tends to become unstable during winding depending on the setting of the control system. It is guessed that the cause is the effect of modeling error due to mechanical problems.

This paper proposes a novel self-reconfigurable robot that is mainly composed of three parts, plastic shell, walking mechanism, magnet connection mechanism. There are two DC motors and two servo motors inside. The two DC motors are used for the walking of the robot, and the two servo motors are used for the relative movement of the two modules after the robots are connected. The robot can complete multiple tasks: the modules can move independently, the modules can move on the water surface after the modules are connected, and they can be connected into a ring to float on the water, etc. Experimental verification of these multiple tasks will follow.

The disasters have occurred frequently in Japan. When conducting rescue activities at a disaster site, the risk of a secondary disaster can be reduced by navigating a rescue robot before the rescue team members and checking the situation at the disaster site. When performing rescue activities, the rescue robot applied to the installed manipulator to open/close valves and remove rubbles. At the disaster site, there are places where water is overflowing due to the rupture of pipes or damage to the tank. Waterproof is required to work in that area. Therefore, in this research, we designed and developed a manipulator with improved waterproof to be mounted on a rescue robot with a hydraulic actuator.

This paper describes a rescue robot equipped with hydraulic actuators and the wheel mechanism with having variable air compressing function. The developing robot is driven by using two traveling units equipped with eight wheels withthe wheel mechanism with having variable air compressing function. By tilting the traveling unit with a hydraulic actuator, the robot can overcome obstacles. The rescue robot can easily travel on uneven terrain by reducing the air pressure in the variable pressure tires. Because reducing the air pressure in tires increases the area of contact to the ground surface. It also describes the control system for the entire robot.

Standard Test Methods(STMs) for small Unmanned Arial Vehicles(sUAVs) developed National Institute of Standard and Technology(NIST) have been widely used for operator training and criterion of deployment and procurement in US. While the sUAV-STMs are mainly used for disaster response in US, the other applications, e.g, aerial photography and infrastructure inspection, are also expected in Japan. Therefore, applicability of NIST sUAV-STM in Japan should be investigated. In this paper, an evaluation exercise of sUAV-STM has been carried including a long outside lane(90 meter length) with sUAV pilots of expert aerial photographers. Based on the measured sUAV-STM data and questionnaires, issues of sUAV-STM application in Japan will be discussed.

In this study, we propose a method to estimate the spring constant k and the bending tendency φ₀ between rigid bodies with high accuracy when estimating the parameters of the pseudo rigid-body model of a continuum robot. Since the spring constant k and the bending tendency φ₀ are deeply related to each other, it is difficult to estimate them simultaneously and with high accuracy. Therefore, we proposed a method to estimate k first by eliminating the term of φ₀, which is not affected by the measured value, and showed that the estimation accuracy can be improved by simulation. It was confirmed that the estimation accuracy of k was within 1%.

Currently, the elderly people have shown a strong desire to purchase ultra-compact mobility vehicle for one or two people, but since the body of the ultra-compact mobility vehicle is small, there is no enough space to install a system that supports steering operations such as power steering. Our research group is equipped with a steer-by-wire system, which is a small steering system, and is studying steering support for drivers. However, since the steering reaction torque that the driver feels appropriate is different for each person, the fundamental consideration of the burden evaluation for the steering reaction torque was conducted in this study.

Ultra-compact electric mobility vehicle is a vehicle with a short wheelbase and has a small turning radius. It has become widespread in situations such as transportation for narrow roads area and car-sharing services, and all kind of people use it. Therefore, the ultra-compact electric mobility vehicle requires a steering system that enables easy and safe driving regardless of age, physique, or driving skill. In our laboratory, we consider that the steering feel and burden are different each of driver, and we have suggested an active steering wheel system that can quantitatively evaluate each driver's steering feeling and burden by using biological information and provide steering support to each. The driver's arm model by inverse kinematics used in robotics was used to calculate the joint moment and measure the muscle activity for the steering burden evaluation, and the fundamental consideration of steering burden evaluation by the steering reaction torque was conducted.

In automated driving in urban areas, the vehicle must be able to operate the safety driving following the traffic rules based on the onboard sensors data and the predefined digital map. When the vehicle approaches an intersection, it is necessary to determine safety based on the status of surrounding traffic participants and traffic lights recognized by the onboard sensors. Especially, the lighting color of traffic lights must be recognized using only the onboard camera. However, there are situations in which it is difficult to recognize the traffic lights based on images alone due to changes in sunlight and weather conditions. In our research project, we have been conducting demonstration experiments of automated driving on Japanese public roads and discussing the limitations of autonomous recognition using onboard sensors. Then, we have developed an image dataset of intersection driving including challenging image conditions for traffic light recognition obtained through our demonstration experiments.

In warehouses, pallets are used for transportation, and overturning and collapsing of loads during transportation are problems. One of causes of cargo collapses is the effect of external forces in transportation, such as centrifugal force caused during turning a corner. In this paper, we propose a method to stabilize the loading during pallet transportation. The proposed method searches a stable loading pattern by a local search method from multiple loading patterns generated by our pattern generation algorithm. To validate the performance of loading stabilization, we comparison the loading patterns using dynamic simulation.]

We have developed a flying robot that hangs a reciprocating saw from a multi-rotor UAV with a wire. In this study, we propose a method of pruning a branch at a place away from the UAV as a work point and experiments aimed at verifying the possibility of pruning work and improving cutting speed and stability. In the experiment, some wood was cut with no force applied and with ducted fans pressed against the object to be cut. In order to measure force and vibration, we attached a force sensor to the fixed UAV and conducted the experiment. We confirmed that the proposed method can be applied to the actual flying robot and succeeded in significantly reducing the cutting time by pressing against the cutting target through experiments.

In recent years, a variety of unmanned aerial vehicles (UAVs) have been developed to replace human tasks, such as carrying cargo and inspecting infrastructures, and have attracted attention in many fields. However, conventional quad-rotor UAVs are forced to change in attitude when moving in the air. Therefore, aircraft that multiple rotors are assigned with an inclination or are tiltable has been developed, but most of them are redundant actuated systems. Therefore, in this study, it aims to develop an aircraft with a 2-DOF tiltable rotors to realize a fully-actuated system, which is capable of independent flight in position and attitude, using a small number of rotors. In this aircraft, rotors being different rotational directions are arranged up and down, one by one, so that it can reduce the power consumption and the risk of failures, because the number of rotors is only two.

Vertical-takeoff-and-landing unmanned aerial vehicles (UAVs) are known as drones and used in a wide range of fields. In general drones, the rotor is fixed to the fuselage, and can only output thrust in one direction, which causes the aircraft to change its attitude when moving. In this research, it is aimed at creating a UAV that can flexibly fly without causing any changes in attitude when navigating. There was a flight limitation due to the total weight of the airframe in a previously studied UAV, which was a tandem-type rotorcraft having 2-DOF tiltable coaxial rotors at front and rear respectively. This research solves this problem and develops a new Osprey-type drone that can fly without such a limitation.

We proposed a long-reach lightweight articulated arm ”Hiryu-III” using weight-compensation with thrusters and a wire-pulley system. This robot uses a wire-pulley system to keep the thruster horizontally, however the elastic deformation of the plastic tensioner causes insufficient rigidity of the pitch axis. We changed the fastening method between the tensioner and the shaft from key and keyway to a hexagonal shaft and hole. Additionally, we conducted experiments to confirm the improvement of the stiffness of the pitch axis. In addition, we conducted experiments with a 3-link prototype, and found a problem of accumulating error of the yaw joint axes. To solve this problem, we proposed a mechanism reversed upside down.

Currently, various researches are being conducted on self-position estimation methods in indoor environments. As a previous research, there is a self-position estimation method that combines an IMU (Inertial Measurement Unit), a monocular camera, and a ground altimeter with an Extended Kalman Filter (EKF). However, in the previous research, the sensor data of the ground altimeter was used as the observed value in the altitude direction, and it is easily affected by the change of topography. In this study, we propose a more robust estimation method for step topography, mainly by adding improvements related to altimeter processing. In addition, ORBSLAM3 is used as the monocular SLAM, and Gazebo, the physics simulator, is used in the experiment.

This paper presents simultaneous stabilization of longitudinal and lateral dynamics considering the input saturation for a flying-wing unmanned aerial vehicle (UAV) only with an elevon that is the control surface combined the functions of the elevator (used for altitude control) and the aileron (used for direction control). A new linear matrix inequality (LMI) based design approach for simultaneous stabilization considering the elevon input saturation is proposed from a nonlinear guaranteed cost control point of view. The three-dimensional flight control simulations demonstrate the utility of the proposed design approach over the existing control approach which separately designs the longitudinal and lateral dynamics and mixes the elevator and aileron commands.

This paper describes a drone simulator that we have developed for autonomous flight verification. Currently, interest in drones is increasing and their applications are expanding, but accidents are also occurring. In order to prevent accidents, it is necessary to thoroughly verify the control law, but there are risks such as crashing when using a real aircraft. Therefore, it is possible to reduce the risk by using a simulator in the testing phase of the control law. In this study, we constructed a simulator using MATLAB/Simulink and Simscape Multibody, and conducted controller adjustment experiment for a single drone and verification of autonomous control laws for multiple drones. From the first experiment, it was found that the simulator could not reproduce the behavior after adjustment. From the second experiment, we verified the control law and clarified its problems.

In this study, we focus on the characteristics of the flap and aim to realize safe automatic takeoff and landing by using flap. We investigated the relationship between the energy efficiency of flap during flight from the relationship between lift during cruising and turning and mechanical energy. Also we consider a method to determine the optimum flap deployment angle according to the speed and attitude of the fixed-wing UAV.

In this paper, we examine the aerodynamic characteristics of a tilt-rotor UAV with several tilt angles near its fixed-wing mode. The lift-to-drag ratios and the battery consumption rates are estimated using CFD (Computational Fluid Dynamics) simulations and a rotor model. The numerical results show that, at a flight speed of 20 m/s, there is no significant reduction in the lift-to-drag ratios and in the battery consumption rates even when the tilt angle is tilted up to about 30 degrees from the fixed-wing mode. We also present the results of flight experiments to compare with the numerical results, and verify that an energy efficient flight can be performed even with the tilt angle of 30 degrees.

In this paper, we report a small jumping mechanism for a lunar exploration robot using shape memory alloy (SMA) actuators. The jumping mechanism have tension springs and toggle linkage with SMA actuators. SMA actuators used for operating the tension springs. In the jumping experiments, we confirmed that the jumping model hoped to a maximum height of 0.34 meters and travel up to 0.5 meters. This indicates that the jumping mechanism using shape memory alloy actuators is useful as a movement mechanism.

Currently, there are an increasing number of space debris in the Earth’s orbit, such as parts from satellites and rockets that are no longer in operation but hold crucial risks of collision with other operational spacecraft. In order to safely approach and capture space debris with an unmanned orbital space robot, it is essential to understand its motion state before the active removal operation. In this paper, we experimentally verify the motion estimation method that uses the Structure from Motion (SfM) algorithm, which is a technique to reconstruct the three-dimensional shape of a target object and the pose of a camera by inputting a group of images with different viewpoints.

This paper proposes a bulldozing procedure using lightweight small robotic vehicles on lunar surface. The procedure mainly addresses a bulldozing task to fill and level ditches (i.e., craters) with considering earth-working characteristics. The characteristics, such as bulldozing efficiency, working efficiency, and energy margin, are evaluated by dynamic simulation of the robot with varying bulldozing length and depth. The simulation result quantitatively provides an efficient bulldozing scheme for filling a rectangular ditch.

The autonomous navigation system is required for planetary exploration rovers to drive to the destination. The conventional navigation systems always perform processing with a large load. Therefore, it takes long time to traverse to the destination. It is necessary for rovers itself to select the behavior modes of navigation according to the environment so that a rover drives faster and efficiently. In this paper, a method is proposed in which the rover autonomously selects the path planning method based on the monocular images. A simulation study is conducted to evaluate the effectiveness of the proposed method.

In human spaceflight activities, robotic task-assistance is anticipated to support safe and effective space exploration, as well as efficient utilization of space environment. Japan Aerospace Exploration Agency (JAXA) Human Spaceflight Division have been pursuing the robotic and automation technology. Cargo handling activities, including stowage of various goods and preparation, closeout task requires crew engagement in present ISS operations. Such tasks consume significant amount of precious crew time. Thus, in this paper prospect for cargo handling task automation is described. Based on the analysis of current operations, tasks anticipated for automation are outlined. Thus, required technology is discussed in 3 aspects. First, inventory management technology which was trade between mass and stowage efficiency between fully automated system. Second, intra-vehicular mobility with cargo transportation capability is discussed by comparing free-flyers and mobile manipulators. Lastly, cargo manipulation is analyzed based on the current crew IF manipulation by robotic manipulator.

In recent years, the number of small spacecraft missions has been increasing rapidly due to reducing spacecraft development costs and diversification of functions. This research deals with autonomous docking between small spacecraft. We propose an optimal control system to perform autonomous docking with high accuracy with limited sensor and actuator resources.

In recent years, skylights, which are vertical holes thought to be the entrances to lava tubes, have been discovered on the lunar surface. Although the robotic exploration inside there is essential for the deep understanding of the lunar geological history, however, conventional wheeled mobile robots are unable to traverse such steep and rocky terrain. For this challenge, We have developed a quadrupedal climbing robot with spine-type grippers that is able to hold such terrain and experimentally demonstrated autonomous operation in a known simulated environment. In order to achieve the autonomous operation in an unknown environment, the robot is required to grip the target rock by its gripper reliably. In this paper, we developed a gripper with a built-in force/torque sensor and IMU and investigated the gripping performance against the different directions of the external force. Based on that, we architect the secure gripping sequence.

For robotic exploration on the Moon or the planets, robots need to collaborate with others in real-time. SLAM which allows a robot to localize itself and map the surrounding area simultaneously is one of the most crucial technologies for autonomous mobile robots. However, most SLAM algorithms were built for only a single robot, and do not provide any solution against collaborative mapping and localization problems. In this paper, we aim to fuse different maps from two robots in real-time without initial correspondences.

In forestry, it is demanded for work vehicles such as forwarders to drive automatically on forest work road, however, devices for self-localization like GNSS and wheel encoder are difficult to use due to its blocked environment. In this study, we propose a method for self-localization in forestry environments. Our method is based on the NDT scan matching that is simplified for real-time computation. Furthermore, it improves the localization accuracy by utilizing reflection intensity of LiDAR to remove environmentaldisturbing objects likeleaves that cause miss matching due to seasonal changes. In the experiment, we verified the self-localization performance of our proposed method in an environment assuming a forest work road.

Currently, the reduction of wetland area has become an important problem in Japan. However, it takes a lot of effort to conduct a human survey. Therefore, in order to improve the efficiency of survey activities of the wetland, a wetland mobile robot that moves without damaging plants as much as possible has been proposed. In this research, we proposed a wetland mobile robot that combines a spiral mechanism and a deployment leg mechanism. The machine design and control method will be examined using the model built in the dynamics simulator. The operation of each mechanism and the entire robot will be verified.

We have been working on the development of autonomous mobile robots for the Arctic sea ice survey system. In this system, the robot is required to be able to run stably on ice and snow, and to run autonomously by highly accurate positioning. In this paper, Russell performance experiments to improve running performance on ice and snow, and route map creation of position information by RTK-GNSS for autonomous driving were performed on Lake Saroma ice in winter. From the experimental results, we clarified the Russell performance during running and achieved the creation of route mapping of location information.

We propose a robot that is initially lightweight and soft but can develop new functions by taking in environmental materials and using them as a structural material. As an example of such a robot, we develop“TsuchiKurai”; a robot that acquires new functions by increasing its stiffness and mass by taking in the surrounding soil and using it as its structural material. As the first step, we devised a soil collection mechanism using a robotic hand wrapped in a bag-like structure. In addition, we developed a prototype of the soil collection mechanism and the stiffness change by soil and conducted experimental verification.

In recent years, there has been a lot of research on forest management systems using UAVs. For the purpose of forest management by UAV in a large-scale forest, we propose a method to create a 3D map of an actual forest, to perform clustering and labeling of the individual trees in the created map, and to add GPS information to the crops.This study proposes a highly accurate management system using UAV, 3D LiDAR, and GPS information.

Inspired by the eating function of living organisms, an energy sustainable robot that recovers dead leaves, which is one of the forest biomasses, and moves continuously with the heat of combustion is proposed. To achieve this, mobility and dead leave collection suitable for uneven terrain, ignition and combustion of dead leaf biomass, charge/discharge of thermal energy, and control of actuators are necessary. In this research, a mechanical model with two large wheels and a hand for collecting dead leaves is developed. Basic power consumption, motion of collecting dead leaves, movement on rough terrain, ignition and burning, energy consumption in continuous operation are evaluated in experiments.

For reducing the labor of weeding in field agriculture, we propose a small weeding robot. We developed a small robot vehicle which has two cameras and a manipulator. The cameras are attached to the robot in a downward direction to capture images of the fields. By applying deep learning to these images, crops and weeds are discriminated in real time. A conveyor belt-like gripper is attached to the end of the manipulator and can uproot weeds growing in intra-row spacing. To proof of our concept, indoor weeding experiments using artificial flowers were conducted. The results show that autonomous weeding by using image recognition was possible.

In order to make precision agriculture more accurate, real-time operations for individual crops is a challenge. It is essential to measure the crop area accurately when approaching individual crops. However, existing instance segmentation methods are computationally expensive, and it is difficult to recognize crops on devices in real-time. In this paper, we develop a crop recognition method for a gripper weeding machine. By creating a high-precision crop map in advance and limiting the image area based on the map, the location and the azimuth of the vehicle from GPS units, we succeeded in the real-time and high-accurate individual crop recognition on Jetson AGX Xavier.