In our research group, we are developing the mobility support equipment to provide children who have difficulty moving on their own from an early stage. The aim is to support children's cognition and emotional development through their own mobile experience. In this study, we have proposed a front wheel drive and a middle wheel drive electric mobility assist devices that can provide the movement experience for handicapped children with a standing position. The prototype device can be operated and moved by a child using a joystick. We also evaluated and compared the performance of two prototype devices. The prototype device was actually used by children who have difficulty moving by themselves and the usefulness of the device was confirmed.

Rehabititation training includes one-handed exercise training and two-handed exercise training. Although many one-handed rehabilitation robots have been developed, the development of two-handed rehabilitation robots is scarce. A two-handed rehabilitation robot is needed to support the paralyzed arm with the non-paralyzed arm. In this paper, we report on the development of mirror symmetric type rehabilitation support system using passive force display as two-handed rehabilitation robot.

In this study, we have developed walking assisting shoes by using elastomer-embedded flexible joint (EEFJ). The EEFJ is a kind of compliant joint, but has non-linear and adjustable mechanical properties. In our project, we have especially focused on the design method suitable for three dimensional motion of human foot and mentality of frail users. In this report, we summarized the results of the interview on walking shoes conducted for frail elderlies in Kituki City of Oita Pref. In addition, we designed shape and structure of frames for dorsiflexion support shoes.

In our research group, we have been researching the Self Transfer Assist Robot that realized the transfer by care receivers themselves. The prototype of the transfer assist robot could transfer from the sitting position to the sitting position. In this research, we tackle a research on "Self Transfer Assist System", that is including the posture change from the supine position to sitting position. This system consists of " leg assistance robot ", " electrical bed with body position changing function", in addition to " transfer assist robot". In this paper, we investigate how “leg assistance robot” receive the legs from “transfer assist robot”. To solve it, we develop “Leg transfer mechanism” with the rack and pinion mechanism and pneumatic soft actuator. By this research, we realized the receiving legs and transferring from the sitting position to lateral position.

In our research group, we have been conducting a research on the Self Transfer Assist Robot that realized the transfer by care recipients themselves. A prototype of the transfer assist robot was able to transfer from the sitting position to the sitting position. In this research, our interest is concentrated on the advanced research on "Self Transfer Assist System", which is capable of the transfer including the posture conversion from the supine position to the sitting position. This posture conversion is carried out by collaborative work with plural robots. This system consists of "the leg part support robot", "the robotic bed with posture changing function", in addition to "the transfer assist robot". In this paper, considering the burden on the care recipients and the interference with the bed, we improved "the upper body holding part ". We create the prototype of the thin upper body holding part that can be snuggled with the upper body and provides the stable transfer assistance.

Chemotactic behavior of a monotrichous bacterium was investigated by the three dimensional measurement. An objective lens was vertically oscillated by the piezoelectric element and the position of the cell was deduced from images in focus. Three dimensional trajectories of cells swimming around an attractant were obtained with the vertical resolution of several micrometers. The turn angle (directional change) distribution, which is one of the characteristics in bacterial chemotaxis, was elucidated. The turn angles in the directional changes from forward to backward swimming are limited to large values (almost 180 degrees), whereas the angles widely range in the directional changes from backward to forward. This tendency is similar to the previous study with the two dimensional measurement.

The paper presents a motion which allows a snake robot to climb a vertical pipe without all round winding. When climbing a pipe, the target shape of the robot consists of a part that touches the pipe and a part that does not touch the pipe. Furthermore, a unique motion of a snake robot called rolling is used. In addition, the problem that the robot shape generated during the movement collapses from the target state is solved by the contact between the robot bodies. With this motion, the robot can climb a pipe with obstacles nearby. The effectiveness of the proposed motion is demonstrated by using a physical simulator.

Walking on the split-belt treadmill is one of the experiment to investigate the gait adaptation. In these days, to make a model that describes the adaptive mechanisms connecting body dynamics and sensor feedback, some researchers have used a leg robot. In our previous studies, we proposed a spinal cat model that is based on a leg controller and Frigon’s spinal cord model using a quadruped robot “Kotetsu” and showed that spinal cat model is not enough to adapt on the split-belt treadmill. In this study, to analyze the motion of spinal cat model, we focus on a translation of contralateral leg loading. As the results of experiments, we show that Kotetsu moves back and forward slightly on the split-belt treadmill, so the spinal cat model cannot adapt on the split-belt treadmill. Spec. and movies of Kotetsu can be seen at http://www.robotlocomotion.kit.ac.jp/kotetsu/index-j.html.

Recent robots have been used in various environments and require complicated movements according to various environment. It is difficult to control the robot in every environmental situation, however, it is possible for a robot to learn the given motion by using machine learning techniques that imitates the infant's motor learning. It is expected that the dynamical properties, such as inertia, reaction forces, internal forces flow could be one of the constraints which support to reduce the exploring space. To investigate the relationship between an effectiveness of machine learning and constraints induced by body-dynamics, we developed the whole systems including tendon-driven legs robot, split-belt treadmill, and hip joint with a six-axis sensor to measure how a multi-legged animal affected itself and the environment during walking. It is confirmed that it was possible to measure the data around the hip joint during walking in experiments.

Confirmation of nesting situation is necessary to preserve animals nesting on trees. However, investigators climbing a tree and approaching the nest may cause a crash or nesting abandonment due to survey pressure. Therefore, a small tree climbing robot that functions as a tree camera is required. We focus on insects that climb trees without bending the trunk because the space for mounting the camera and the gripping force are large relative to the body size. Insect legs consist of coxa, trochanter, femur, tibia, and tarsus, and the bending of tarsus contributes to tree gripping force. We modeled tarsus and designed the underactuated leg based on tarsus. The developed leg can be driven like tarsus and the tarsus part of the leg can be bent 49deg.

We conducted the motion analysis of polychaete worm, and developed the omni-directional aquatic multi-link propulsion mechanism. Movement of fins of the omni-directional aquatic multi-link mechanism was based on the motion of parapodia of polychaete worm. In this study, the width of fin corresponding to the length of parapodia were changed. We used soft elastic fins to deform easily when adjacent fins were interfered for greater width of fins. In this report, we changed the timing and direction of the movement of the fin and examined the effect on the propulsion speed.

Human subtalar joint has the functions of suppressing excessive rotation of body and impact of landing during walking. The purpose of this study is to realize these functions intothe robotic foot. We constructed a robot foot that imitated the rotation axis of the subtalar joint and performed a walking experiment by a walking simulator. The experimental results showed the vertical free moment and the floor reaction force were reduced. Our results confirmed that the rotation axis of subtalar joint could suppress rotation and impact at initial contact.

Pill bugs perform a defensive motion to protect the abdomen by momentarily bending of the body against slight stimuli. Although the development of robots that exert such instantaneous power has been actively developed, all require precise control system design, and there is a delay in response due to electrical feedback control. In this study, we propose the mechanical structure realize instantaneous bending behavior Pill bugs. For the structure we adopted a tensegrity structure as a method of manufacturing a robot to realize it. By adopting a tensegrity structure as a robot body that can exhibit whole-body perception and instantaneous behavior, it was confirmed that the structure realized a movement close to a Pill bug with lightweight members, and without electrical signals.

Penguins exploit not only their wings but also the head to achieve agile locomotor capabilities, such as rapid turning and acceleration. Although mimetics way of penguins’ swimming would lead to agile underwater robots, it is still unclear how penguins change their locomotor patterns depending on situations. As a first step to understanding the underlying control mechanism, our previous study has reported different wing-head coordination manners between slow and fast swimming speed through behavior observations and kinematic analyses. In the present study, we developed a simple two-dimensional physical simulator and evaluated the relationship between wing-head coordination patterns and locomotor performance in penguins’ swimming. Simulation results show that the coordination manner between wings and head exhibited by actual penguins achieves the fastest speeds than other coordination manners.

Snakes are code-shaped without limbs and can realize multiple locomotion modes: sliding, swimming, climbing, and gliding. Gliding snakes undergo aerial locomotion by using its entire body as a flattened, moving wing, which is constantly reconfiguring throughout gliding flight. Aerial locomotion in gliding snakes is obviously distinct from any other type of gliding with high amplitude body undulations visually dominating the behavior. This paper presents concept of a mechanical model named NURSE-I inspired from gliding snake and development of the first experimental model. Then, the gliding flight experiments are reported to verify the basic performance of the constructed experimental model.

In this paper, we proposed a variable capture-angle sampling system for a chemical plume tracing (CPT) robot. We designed a variable capture-angle sampling system and implemented it to an autonomous mobile robot. We also proposed a heuristic control of the angle during CPT. Through verifications using ethanol as the target chemical, we demonstrated the variable capture-angle sampling ability of the proposed system. Additionally, we conducted CPT experiments with the heuristic control to change the capture angle of the system depending on intervals of chemical senses. The results indicated that the proposed system potentially improves the efficiency of CPT by autonomous individual robot.

The recognition of the environment is significant for humanoids to realize advanced tasks. In this study, we develop an ear unit for human mimetic humanoids to realize human-like sound recognition of the surroundings. This ear unit is composed of outer ear structures, MEMS microphone boards and a FPGA sound processing board. We conduct two evaluation experiments of this ear unit. First one is a confirmation experiment of frequency response fluctuation by changing the pitch angle of the head and we acquired filters which depend on sound source direction by introducing outer ear structure. Second experiment is an estimation experiment of horizontal sound source direction based on Jeffress model[1] and we succeeded to estimate sound source direction in front.

We have been investigating how to use behavior-controlled Daphnia as a living micromachine. The control of Daphnia was done by applying their phototaxes. We successfully controlled the behavior of Daphnia using phototaxis. However, Daphnia has millimeter size body and it is too big for micromachines. Therefore, we have been investigating for the other applications such as for the entertainment field. We developed a game that guides Daphnia to 10 places in the previous report. There were many flaws in the behavior control system used in this game and we had problem of not being able to accurately control the behavior of Daphnia. We examined rebuilding the behavior control system using LCD screen and developed a new system. We also developed a traveling method in the large game world using small motion control pool. Based on them, we developed a computer game to compete the Daphnia’s maneuverability.

In this study, the controllability of the behavior of jellyfish was investigated by applying their phototaxis. The displays of jellyfishes are very popular in the aquarium now. If we can control jellyfish, we can make unique amusement system using this technique. We investigated moon jellyfish (Aurelia aurita) and papuan jellyfish (Mastigias papua). Both species have strong positive orientation phototaxis to the blue light. Therefore, we can control their behavior by gathering them to the irradiated light. We succeeded to control a moon jellyfish to every four corner of the rectangular pool automatically using computer and image processing technique. The swimming performance of jellyfishes, however, are very weak and they need continuous stream to keep them in the middle water. Therefore, if they are in the still water, they soon lie down on the bottom surface. It means that we have to control in the turbulent water and this condition makes the controllability of jellyfish worse.

We have been developing motor-tendon-joint series connection type biomimetic actuators to make robots act jumping, running and dancing. In this paper, in order to use the tendon structure effectively, we designed new tendon structure In order to increase the energy recovery rate, we investigated effects of shape of the tendon structure. We found the matching of the natural frequency of the tendon rubber jumping timing was important.

Myriapoda, having multitudes of legs and an elongated body, can dexterously travel on natural environments. Myriapod locomotion has an advantage over wheeled and tracked vehicles on a rough terrain, because each leg can discretely contact the ground at several points. This study aims at developing a dynamic model of the myriapod robot using frictional force and elucidating how it locomotes.

The musculoskeletal morphology of a human being should have a rational structure for the function of bipedal walking. If we recognize the human walking function, then we can reconstruct it by artificial materials with robotic technology. In this study, the muscles in the lower extremities, including the biarticular structure, were replaced by the simple springs-cable element, and three posterior muscles were actuated by electrical motors as a robotic system. The musculoskeletal walking simulator successfully reproduced human-like walking trajectory from the viewpoint of joint angles.

This research focuses on “gracefulness” and aim for the extraction and modeling of the graceful motion feature using classical dance. We are performing grace feature extraction by extracting S-shaped curves from hand trajectories. We considered that the S-shaped trajectories due to hand trajectories include both those created by changes in position and orientation of dancers and those created by the dancer’s arm movements. In this paper, we report on the grace features extraction by converting motion data acquired by motion capture into data in which only limbs move, and extracting frames in which arms move. We compared the correlation between hand trajectory shapes of motion data acquired by motion capture and motion data moving the arms, and impression evaluation results. As a result, we showed that using only the motion of moving the arm resulted in feature quantity that matched impression evaluations.

We have developed a method that enables a robot to follow a person robustly using a laser range finder. In this paper, we propose a control method of a robot that moves side-by-side with a person. Specifically, we develop a method for a robot that moves on the right side of a person. Next, we develop a method of selecting and following either left or right side of a person. Finally, we develop a method that allows the user to freely select the left and right positions. Experiments following the S-shaped path confirmed that the proposed methods worked.

Empathy is very important in the enhancement of rapport, and broadly divided into emotional empathy and cognitive empathy. However, there are few approaches to expressing emotional empathy that can be understood intuitively. In this study, we designed and developed a teary-eyed for displaying emotional empathy during human-robot interaction. This robot mimics the human lacrimal structure, and can generate teary eye by controlling the inflow of water. The effectiveness of the teary-eyed was confirmed experimentally.

This work presents a framework for the kinetics and kinematics estimation of the human knee joints in natural movements. In this study, we employed the long short-term memory (LSTM), a special recurrent neural network (RNN) architecture, as an estimation algorithm by using electromyography (EMG) features. The results suggest a potential application for online prediction of joint kinetics and kinematics without ground force measurements.

This paper describes a novel approach toward automatic annotation of infant’s behaviors following the collection of action descriptions in the crowdsourcing. Videos containing actions are publicly opened, and are manually annotated by internet users in the crowdsourcing platform. This results in a large training dataset of infant’s actions and their relavant descriptions. Relation between the actions and their descriptions are extracted by a probabilistic framework with two modules; semantic module and syntax module. The semantic module trains the association of the words from the action, and the syntax module trains the word sequence with grammatical consistency. This framework allows for automatic conversion of action observation to annotations. Our proposed approach was tested on actions performed by an infant and its validity was demonstrated.

Crutches are often used to relieve a load on the lower limbs. When an inexperienced patient of using crutches walks with crutches, the walk motion sometimes becomes irregular and the patient tends to lose the body balance. Therefore, in this research, a crutch walk training system with AR technology was proposed focusing on an obstacle avoidance. In this paper, a training space of crutch walks where AR markers were attached on the walkway was developed. When a crutch user who wears a see-through HMD walks, and the camera of the HMD detects an AR marker, a virtual obstacle is presented at the place of the AR marker on the HMD. The experiment participant was directed to avoid the virtual obstacle, and the accelerations of the user’s head were measured. As a result, it was found that the accelerations of the crutch user’s head changed irregularly when a virtual obstacle was presented in front of the crutch user on the HMD. It is expected that by repeating such a training, the inexperienced crutch user will get used to a proper walk method with crutches.

The final goal of this study is to quantify claudication focused on the beauty of appearance and propose a new rehabilitation method based on the evaluation. This study norms the model walk as a beautiful walking motion. We try to quantify the walking motion using the similarity to this norm. First, we create a Ruled Surface from silhouette images of the walking and represent the walking motion as a parametric surface. We extracted the magnitude and distribution of the horizontal swing as walking features using this Ruled Surface, and it was found that there is a characteristic difference between the model and general male walking.

For work load reduction of the agriculture, remote control of the agricultural machinery becomes important. This study uses VR device to confirm the situation around the agricultural machinery when agricultural machinery is remote-controled. And a system performing instructions to agricultural machinery by a gesture is suggested.

In this paper, we will discuss the shape feedback control for an isometric closed-curve robot. The shape feedback control is formulated based on a curvature flow equation, which is a partial differential equation similar to the heat transfer equation. The authors have proposed curvature flow based shape feedback control for surface robots, but surface robots that can be deformed arbitrarily have not been realized yet. Therefore the proposed feedback control is investigated using an isometric closed-curve robot, which is a system similar to a surface robot. In this paper, we will show that the proposed controller has the ability to converge the robot’s shape from initial states to target states.

AI-like data driven control ‘V-Tiger’ (Virtual Time-response based Iterative Gain Evaluation and Redesign) calculates a virtual time response when a certain controller K is used based on one experimental data, and sets K to reduce the settling time and overshoot. Also, low-impact active vibration controlled casters that utilize center of percussion are attracting attention to realize low-impact, low-vibration carts. However, this system is a one-input, two-output system, and it takes time and effort to adjust the five control gains of the controller. Therefore, establishing a systematic method for determining the optimal gain is mentioned as a remaining issue. This paper expands V-Tiger for multi-output system and applies to the active caster utilizing center of percussion.

Since Artificial Intelligence (AI) has won over human pros such as Chess, Shogi and Go, expectations for AI have been increasing dramatically. One of the reasons why AI has developed so much is the tremendous increase in processing speed of computers, which makes it possible to virtually repeat simulated competitions such as Othello, Shogi, Go and so on in the computer very fast. Finally, AI has gained strength over human pros. Also in control engineering, if gain tuning experiments of controllers can be virtually performed in a computer, it can be expected to dramatically improve control performance with an AI-like approach. This paper practicalizes the method called ‘Virtual Time-response based Iterative Gain Evaluation and Redesign’ (V-Tiger) which iterates 1) to calculate virtual time responses of the closed loop system when a certain controller is inserted based on one-shot experimental data, 2) to measure the overshoot and settling time from the virtual time responses, and 3) to evaluate and redesign the controller gain considering the stability margin.

AI-like data driven control ‘V-Tiger’ (Virtual Time-response based Iterative Gain Evaluation and Redesign) calculates a virtual time response when a certain controller K is used based on one experimental data, and sets K to reduce the settling time and overshoot. Also, low-impact active vibration controlled casters that utilize center of percussion are attracting attention to realize low-impact, lowvibration carts. However, this system is a one-input, two-output system, and it takes time and effort to adjust the five control gains of the controller. Therefore, establishing a systematic method for determining the optimal gain is mentioned as a remaining issue. This paper expands V-Tiger for multi-output system and applies to the active caster utilizing center of percussion.

We propose a design method of a recently developed Unknown-Input Estimator (UIE) using Artificial Bee Colony (ABC) Algorithm. The design of the UIE is based on the optimal control theory, and two weight matrices must be specified to compute two estimator gains. Therefore, we devise an appropriate objective function to estimate disturbance utilized in the ABC algorithm. The designed UIE is applied to the position control of a three-link manipulator, and the experimental results show that the proposed approach is effective for the design of the UIE.

Many researchers studied path planning and trajectory tracking of the tractor-trailer system. Most studies used a differential drive vehicle for a tractor. However, differential drive vehicles are stuck in some cases due to restrictions on their travel directions. In this paper, we solved that problem with the use of an omni-directional tractor, a kind of mobile robot being able to move every direction. As a trailer, a differential drive vehicle is used, which has no actuator. We control the tractor-trailer system by model predictive control based on its kinematic model and consider both path following and obstacle avoidance. Simulation results show its performance of the proposed tractor-trailer system.

In this paper, we propose a grasping strategy that grasps an approaching object whose position and shape are unknown using a robot hand and a robot arm equipped with a proximity sensor at the fingertip. As a grasping strategy, pregrasping, following, preshaping, and grasping motion necessary for grasping the moving object were created based on a torque control base, and the target joint torque output at the time of each motion was weighted and added. The robot hand and arm movement is generated by the output. It is thought that by gradually changing the weight of each motion in a series of grip control, seamless motion switching is possible, and an appropriate grasping motion is generated.

This paper presents a method to smoothly switch model predictive controllers which are designed for adaptive-cruise-control(ACC)/lane-keep and lane-change tasks. These controllers are defined by different state-spaces, constraints and cost functions. Thus, it can be occurred to violate the constraint when switching controllers. In our proposed method, we interpose an intermediate mode between two controllers to avoid such infeasibility. We demonstrate the effectiveness of the proposed method by a numerical simulation of the mode transition between Adaptive-Cruise-Control/Lane-Keep mode and Lane-Change mode.

Recently, imitation learning has attracted attention as the method for robots to execute tasks in response to environmental changes. Previously, imitation learning using bilateral control showed successful results for tasks requiring force adjustment. However, these studies had difficulty in executing the tasks requiring long-term inference and their verification is not sufficient. Robots must perform long-term tasks for more general tasks. In this study, the performance of the long-term task with a deep neural network (DNN) model was verified. The deep neural network model was expected to extract more long-term features. The validation was performed on the task of writing the letters “ABC”, and the robot could execute the task using a DNN model.

In this study, the open source software Neural Network Console was used to control a Mecanum AGV with omnidirectional capability. First, we measured the location of the corridor using images and magnetic sensor values and obtained good results. Next, using the image information, magnetic sensor information, start and goal information as input, the AGV operation amount was learned. As a result, it was confirmed that the control value of AGV can be estimated directly from the input data.

PID controller is well used for mechatronics control and robot control in industrial use, because of its usability and high performance. There may be several ways to transform a continuous PID controller to a digital PID controller, i.e. Tustin transform and backward difference. However, it is difficult to compare controllers with different structure, because both controllers can achieve any performance in ideal theoretical situation. In practical situation, there may be some limitation of performance in practical situation, which comes from the calculation delay in digital systems. The Limited Pole Placement (LPP) method, which is proposed by the author, can place “assigned poles” accurately even they have delay, and can show the performance limit of the digital systems with positions of “determined poles”. Tustin transform and backward difference are compared using LPP method.

This paper describes planning a high acceleration trajectory under the constraints of a contact force. There are some cases of holding an object with a weak gripping force in order not to hurt the conveyed object. In this case, when the contact force exceeds the limits and the conveyed object may drop. It is important to model contact force and to plan trajectory under constraints of the contact force. We propose a model of contact force and algorithm to search a trajectory that is in the minimal time and is under the constraints of the const force. We demonstrate high acceleration transfer using a manipulator.

The performance of the propeller propulsion system depends on the rotor thrust control performance. Therefore, accurate thrust control is important for high-accuracy control of the airframe. Until now, various thrust control methods have been proposed. In the proposal of a thrust controller that is regarded as a constant thrust coefficient or a function of the advance ratio, a steady-state deviation occurs in the thrust response when the airspeed fluctuates. Therefore, we propose a gain-scheduled rotor thrust controller that uses the thrust coefficient as a two-variable function of airspeed and angular velocity. The conventional method and the proposed method were verified by simulation and experiment.

Assessment of water quality in aquaculture fields requires fine-scale observation over wide area. A quad-maran unmanned vessel was developed as a new type of unmanned ship for automatic collection of various data on water quality at aquaculture fields. In this study, we develop a new trimaran experimental vessel, which is smaller and lighter than the quad-maran unmanned vessel. The newly developed trimaran vessel has three individually turnable hulls and can be used as a tested for evaluating various control and measurement systems on multi-hull robot vessels. This paper reports an overview of the system configuration and functions of the trimaran experimental vessel and shows the results of sea trial in Nanao bay, Ishikawa Pref.

General crawler robots for nuclear disaster cannot move in lateral direction against the robot travelling direction, which causes difficulty in avoiding dropouts and obstacles on stairs. In previous study, our laboratory developed a mecanum crank, which consists of crank wheels and mecanum wheels, and confirmed its omnidirectional locomotion on a flat and uneven terrain. Owing to the interference between the robot frame and the stairs, the robot could not be traversed on the stairs. Furthermore, non-contact of several mecanum wheels on rough terrain prevented the lateral movement. This paper proposes the following additional mechanisms as well as the frame renovation for the mecanum crank the swing mechanism and the roller grouser. In the experiments, we confirmed the robot locomotion on stairs as well as on a flat terrain. Furthermore, we evaluated for swing mechanism and roller grouser.

We manufactured a rough terrain mobile robot to fix the tether quasi-statically to the external environment by installing a long-reach arm on the mobile body. We use an elastic telescopic arm. This arm uses pipes with lower rigidity compared to the conventional telescopic arm, so it can be bent using a rope. Moreover, we can control the tip while avoiding obstacles in the direction of extension. In addition, since the tip can be directly controlled with a rope, the tip can be controlled precisely even when the arm becomes longer. In this paper, we constructed the robot and confirmed the realization of elemental technology.

This manuscript describes the rolling locomotion of a polyhedral robot. The outer shell of the prototype is made of an expanded icosidodecahedron, which has four types of contact surfaces. This robot has an eccentric motor inside. This robot rolls by the reaction force of the centripetal force when the eccentric motor is rotated. This manual clarifies the fundamental physical characteristics, the rotation angle against a slope, of the prototype. When the rotation angle is small, the prototype can roll easily from the contacting surface. Moreover, the experimental results reveal that the translation speed of the prototype increases when the angular velocity of the motor of the internal mechanism is set appropriately.

In order to inspect social infrastructures such as bridge or tunnel, the wall-climbing robots have been developed based on Universal Vacuum Gripper, in short UVG. UVG adhere to uneven surface by the deformable rip part. The robots has simple structure composed of two legs with UVG and two rotational DOFs, and achieved the uneven wall climbing and move along ceiling. However, since this robot had only one adhesion point when climbing the wall, there was a risk of falling due to a failure of the vacuum pump or insufficient negative pressure on the adhesion surface. In addition, the position control of the adhesion point was difficult because of its simple structure.

In this research, in order to avoid the falling and control the position and angle of the adhesion point, we developed the four-legged wall-climbing robot based on coaxially arranged 3 DOFs legs. We evaluated the developed robot in uneven surface walls.

The grass cutting work is a great burden for farmers in mountainous area because engaging grass cutting on steep slopes is one of dangerous works owing to unstable postures and precarious footholds. In this paper, the locomotion abilities that are required for mobile robots on steep slope for grass cutting works are discussed to enhance the development of agricultural robots in mountainous area. The grass cutting robots are necessary to work on a slope with an angle up to 80 degrees with steady attitudes. It is difficult for ordinary mobile robots to keep stable attitude on steep slopes due to slipping and falling. We analyzed previous researches on grass cutting robots on steep slopes and its ability of locomotion. We suggested possibility for using multi-copter type mobile robot into grass cutting works.

In this paper, a stair-descent control method for a stair-climbing single-wheeled inverted pendulum robot is proposed. The objective of the control method is to ensure that the robot can descend a staircase sequentially without losing stability. The experimental results confirm that the robot successfully descends a staircase consisting of four steps with a riser height and a tread depth of 12–13 cm and 39 cm, respectively.