Collaborative robots are one of the useful workforces in the future because they can work with people in the same place. Proximity and tactile sensors play an important role in ensuring the safety of collaborative robots. In this study, we propose a combination sensor for collaborative robots with a time-of-flight (ToF) sensor and self-capacitance proximity and tactile sensor. The ToF sensor can detect the distance to an object in the non-contact range. However, the detection accuracy and X–Y measurement range are reduced at short distances. Self-capacitance proximity and tactile sensing can detect objects at short distances and detect contact conditions. The prototype sensor detected the distance of objects less than 300 mm, and detected the pressure and the position on contact. We think that the proposed sensor can be used as a proximity and tactile sensor for collaborative robots.

In this study, we developed a system to analyze the skating motion of a small biped robot using 3D point cloud obtained by depth camera. To recognize the motion of the robot, four LED feature points were installed on the body. By measuring the positions of the feature points, the position and orientation of the robot were calculated. In addition, we control the luminous pattern of a Red LED feature point fixed on the body in order to record the start time of skating motion. Within 2.0m, we were able to measure the centroid and attitude angle when skating at 30.0fps. It became possible to evaluate the skating posture.

In this paper, we propose a method for estimating the posture and measuring the contact force of roller brushes during painting walls. The posture of the roller brushes to be estimated is the three-dimensional position of the roller section, its rotation on the coating surface, and the angle between the roller brushes handle and the coating surface. The contact force is between the roller brushes and the painted surface. From the verification experiments, we confirmed that the proposed method can estimate and measure these parameters.

In this study, we are developing a method for automatically calculating the next measurement point in order to obtain the shape of the entire object closely in 3D laser measurement. The proposed method is to obtain the whole object high densely by focusing on the low-density part of the point cloud. The next points to be measured are calculated by calculating the relative densities of the points that can be obtained from the distance to the object and the angle of incidence of the laser. In this paper, we show the index which is considered as high density in the field of surveying, and the result of applying the index to a 3D model by simulating the measurement with LiDAR.

This paper discusses how to calibrate our developed real-world clicker using machine learning technologies. Our real-world clicker is a system that can measure the 3D coordinate value of a laser spot. A TOF type laser distance sensor is mounted on a pan tilt actuator in which the 3D coordinate value can be calculated from the pan angle and the tilt angle and the laser beam length. Inherently the measurement accuracy is reduced due to assembly error of the actuator and the laser sensor. We propose a method to correct the error by calibration using machine learning.

In recent years, rescue robots have been expected to play an active role in disaster sites. However, it is difficult to judge a situation from only sensor images of unknown environment. In this paper, three-dimensional presentation system of environmental information by integrating environmental attributes obtained from different types of special sensors and depth information is proposed. In the proposed method, the sensor system uses a polarization camera and an infrared camera. The system can be calibrated by the proposed method, and the integrated environmental semantic information is obtained as point cloud without spatial gaps.

In the medical and biological fields, three-dimensional measurement in a minute space is required, especially when the observation object is a moving object, and it is necessary to track the object at high speed. In addition, fluorescence microscope used in biological observation is an important technique for observing living organisms, as it enables limited illumination and observation of specific areas to be observed. In this study, we developed a prototype system for high-speed 3D tracking in fluorescence observation by combining a TAG lens, which is a high-speed focus tunable lens, and an image intensifier, and achieved millisecond-order 3D tracking of fluorescent beads.

This paper describes a biaxial angular acceleration sensor utilizing the inertial force in spiral channels and MEMS (Micro Electro Mechanical Systems) differential pressure sensors. The sensor unit of the proposed device consists of two rotational-symmetric spiral channels and piezoresistive cantilevers. When angular acceleration is applied, inertial force is generated in the fluid inside the spiral channel. This inertial force deforms the cantilever, causing the resistance change. Therefore, the applied angular acceleration can be calculated from the resistance change. Also, the rotational-symmetric spiral channels cancel the non-target axis sensitivity. By placing two such sensor units at orthogonally, angular acceleration can be measured simultaneously in biaxial axes. Our proposed device responded with almost constant sensitivity from 1 to 15 Hz, and the non-target axis sensitivity was approximately 1/300 of the target axis sensitivity.

In recent years, drone technology has been dramatically developed, and agriculture, infrastructure, and delivery are one of the applications. It is desired to measure airflow condition during drone maneuvering from a viewpoint of safety. Here we propose a sphere shaped two dimensional (2D) airflow sensor with Microelectromechanical systems (MEMS) differential pressure (DP) sensors that can be attached to a drone. The sensor size and weight are sufficiently small and light, respectively. On the surface of the sphere, there are some inlets which give differential pressure to DP sensors located inside of the sphere structure. From the differential pressures detected by the DP sensors, airflow velocity and direction are measured. In this paper, we designed and fabricated a 2D airflow sensor, and investigated the response of the fabricated sensor.

We developed a micronozzle array integrated with a pneumatic channel that can isolate, encapsulate, and eject single cells in droplets with high throughput for applications such as 3D cell assembly and single cell level co-culture. The device consists of 3 layers namely Pneumatic channel, Variable PDMS Membrane, and Cell Channel. Cell solution is introduced into the Cell channel and cells are trapped in specific chambers. These cells can then be ejected by varying the pressure in the Pneumatic channel. Through initial experimentation, we achieved a cell trapping efficiency of 50% (Single cells trapped in 5/10 chambers of the nozzle array) and cell capture rate of 6%. This study is aimed at optimizing the cell trapping phenomena in the micronozzle array to improve the cell trapping and yield better results. Cell were trapped at an efficiency of 90% after making changes in the cell channel design and were captured at a rate of 25%.

This paper reports a measurement system for angle-of-incidence characteristics of a plasmonic photodetector using a strain gauge on a metal cantilever. The photodetector is installed on the metal cantilever equipped with the strain gauge at the root of the cantilever. By deforming the cantilever, the angle of incidence of the photodetector was scanned. This angle of incidence was calculated by measuring the strain on the metal cantilever. The photocurrent signal from the photodetector and the strain signal from the strain gauge were simultaneously measured using an oscilloscope while the cantilever was being deformed. The simultaneous measurement of the two signals produced similar result to the one obtained by a conventional measurement using a rotating stage, indicating that the proposed structure is capable of simultaneous measurement of photocurrent and angle of incidence.

A method of combining an coded exposure camera and a TAG lens with a microscope has been reported as a method of measuring 3D information in micro-space. It has been shown that high-speed 3D information measurement is possible by measuring images at multiple predetermined focal points at high speed. In this study, we propose a new autofocusing method to deal with the motion blur problem by adjusting the exposure based on this high-speed 3D microscope system, and report the effectiveness of the proposed autofocusing method from the experimental results.

This paper reports a current-sensing surface plasmon resonance (SPR) detector that operates in pure water using perpendicular incidence of light. The SPR detector has a gold diffraction grating on n-Si. The absorbed photon energy is converted into photocurrent by the Schottky barrier formed at the Au/n-Si interface, and can be detected as photocurrent. Using simulations, we searched for a grating structure that shows efficient light absorption in the near-infrared region. Based on the simulation, we fabricated the device and measured its reflectance and photoresponse against perpendicular incidence of light. The reflectance measurement showed similar spectrum as simulated. In the photocurrent detection experiment, photocurrent was generated at a wavelength of 1100 nm. Therefore, it was verified that the device can induce SPR and detect photocurrent by perpendicular incidence in the near-infrared region.

In recent years, industrial application development of MR (Mixed Reality) technology using smart glasses has been actively carried out. However, in order to share information with multiple people, it is necessary to prepare smart glasses for each person, which poses a problem in terms of introduction cost. If a transparent, large-area 3D display is realized at low cost, MR display will be possible by superimposing a stereoscopic image on a real object. Therefore, we are developing brand-new method for MR display of full-parallax 3D images by controlling the directivity of light with a transparent holographic screen. In this paper, we describe the basic principle and the method of generating a projected image format that combines parallax image groups. In addition, we report the results of the exposure trial of the holographic screen and the results of the principle demonstration experiment of 3D image display.

We studied adhesion force of a suction cup which can be changed its rigidity during adhesive process. We figured out suction cup with high Young's modulus can generate high adhesion force, but on the other hand suction cup with high Young's modulus needs high pushing force. Therefore, we designed Young’s modulus of a suction cup to be low during pushing and high during pulling. Shape memory polymer which can change Young's modulus with temperature is used for changing the rigidity of suction cup. Then, we measured adhesion force for different temperature of shape memory polymer at the time of pulling. As a result, we achieved a 22% increase in adhesion force by changing rigidity of suction cup.

This paper reports the development of an acoustic sensor using a Helmholtz resonator (HR) array and a Micro Electro Mechanical Systems (MEMS) piezoresistive cantilever-type differential pressure (DP) sensor element. We designed several arrays with different resonant frequencies (RFs), which are similar to the mechanical RF of the cantilever. Acoustic waves that pass through the HR array are amplified around the RF, thus, the DP sensor responds with high sensitivity around the RF. In the experimental result, all devices with the HR responded to the acoustic waves with approximately 4 times higher sensitivity than the device without the HR at the RF of each HR.

In recent years, multi-material stereolithography, which enables the fabrication of highly functional and high-definition 3-dimensional parts, has attracted much attention. However, most of the conventional systems have a problem that it is difficult to use highly viscous materials. In this study, we have improved the cleaning system of our lab-made multi-material stereolithography system using multiple droplets in order to completely clean high viscosity materials. Using this system, we succeeded in making 3D silica structures using highly viscous silica slurry. In addition, we used two types of silica slurries doped with two different metal ions, Eu and Cu, and created a multi-material silica glass structure without material contamination. This method is expected to be applied to various functional devices such as optical and microfluidic devices.

In this study, the method of producing the micro lens array by the MEMS process is proposed. Since the micro lens array is placed on each pixel of the liquid crystal display (LCD), the micro lens array gives directivity in an arbitrary direction for each LCD pixel. Since it is difficult to manufacture the micro lens array designed for each LCD pixel by conventional technology, the micro lens array is produced by using silicon sheets etched by deep reactive ion etching (deep RIE) as a casting mold. The manufactured micro lens array was measured and evaluated using a laser microscope whether it has optically accurate.

Robotic automation of agricultural work is needed to reduce agricultural labor and improve efficiency. In this study, we focused on harvesting tomatoes, which have a high labor load on workers. Tomato recognition is a particularly important factor in the automatic harvesting of tomatoes by robots. Two points were extracted from the three-dimensional point cloud of the fruit cluster acquired from an RGB-D camera. And the intersection of the two points was voted to study the estimation method and segmentation method of the center point of tomato. We compared recognition accuracy and processing speed according to the position taken by the two normal vectors.

This paper provides a case study of AI education at an underwater robotics competition held in 2019. The participants were asked to break three colored balloons in a tank by their robots, at the competition.

Seven teams participated in the competition. Two teams were adults, three teams were university students, one team was from a technical college and one team was from a high school. Three teams were able to break at least one balloon. Two teams used AI. The winner was not the one that used AI. From the above, we found that underwater robot competitions can be used for AI education, but the rules need to be carefully considered.

Team CIT Brains is developing autonomous soccer robots for the RoboCup Humanoid League from 2006 to 2021. We have developed five types of robots for the KidSize League and are repeatedly testing them on the RoboCup. Currently, we are developing an open platform robot GankenKun, and are publishing its design data. In the team, inexperienced students gain practical experience through the operation and development of the robots. Many of the graduate students are working on this project as research. By applying the results of their researches to the robots, we have made the robots more robust and intelligent. In this paper, we introduce the development of humanoids in the team CIT Brains. We also describe how we have developed and passed on the technology within the team.

In the realization of symbiosis between humans and robots, how robots can understand human behavior is an important issue. In the soccer robot, it is possible to show that human-robot symbiosis in a dynamic environment is possible if humans and robots share the same understanding of behavior. In this paper, a robot in a simulator that behaves according to a certain rule is analyzed using a self-organizing map, and the rule is mapped to two dimensions as a feature map and evaluated. By using this map, a system that selects the next action from the position of each player is completed. This system can also be adapted to humans who act according to certain action decision rules.

Educational change has been required, such as the OECD presenting the abilities to survive in a changing future society as "Key Competencies." Since 2003, we have been practicing "Robot Science Education" to solve social problems through robot development. This education is linked to the philosophy of STEAM education, and it focuses not only on robot development but also on approaches to social problems. To verify the effectiveness of this education, we surveyed the students who challenged the robot contests about the abilities they acquired. This study concludes that the Robot Science Education is effective in developing students' abilities for "Key Competencies." Amid this rapidly changing society, the need for Robot Science Education will increase.

Reentry agriculture in Japan is faced with serious problems, which are the shortage of workers, the falling birthrate and the aging population, global warming and natural deserter and so on. The solution to these problems is to build a farm management system using the Internet of Things (IoT) and to automated farm work using autonomous robot. Every year, the Tomato harvesting competition is held by Kyushu Institute of Technology in December with the goal of developing agri-robots.

In this research, we investigate an autonomous fruit harvesting robot with the aim of reducing the labor burden of farmers. In this paper, we report on hardware improvements, development of autonomous algorithms, and experiments result using the classical PID control as a control algorithm for the rail linear motion.

In recent years, filter cloths and bag filters have been installed in the exhaust channels of waste incineration facilities (hereinafter referred to as incinerators) to be constructed. This is because incineration of municipal waste and industrial waste does not release dioxins and heavy metals into the atmosphere. If the incinerator is operated for a long time, the filter cloth and bag filter will be clogged. Filter cloths and bag filters should be cleaned regularly. As for the cleaning method, workers take dustproof measures when the incinerator is stopped, and the incinerator is dropped to the incinerator side using the air from the air hose. Workers have a poor foothold and are not hygienic, so there is an urgent need to improve their safety and hygiene. This time, we have developed a robot that can fully automatically clean filter cloths and bag filters.

In recent years, waste incineration facilities (hereinafter referred to as incinerators) to be constructed are equipped with boilers to generate steam from the heat of burning discarded garbage and recover energy to generate electricity. However, if the operation is continued, the boiler in the incinerator will accumulate ash and reduce the energy recovery efficiency, so it is necessary to clean and inspect it regularly. At that time, a worker enters the incinerator and cleans it, and inspects the boiler pipe for any unpolished residue. There is ash containing dioxins in the incinerator, and the cleaning workers take careful measures to prevent dust. In terms of work efficiency and safety, it is an urgent task to perform cleaning and inspection work with a robot. We created a robot that moves between narrow pipes with an emphasis on heat exchange.

A mobile robot has been developed that treats household waste with biomass and automatically distributes it to the field. The entire system is controlled by a webcam and a general-purpose microcomputer Raspberry Pi. The mechanical system was fabricated with a 3D printer. The overall size is about 50cmx30cmx20cm. The capacity of the biomass tank is about 20L. The tire diameter is about 130mm. In this paper, we report on a biomass processing system that effectively discharges biomass to the outside, a drive system that moves the field on four wheels, and an automatic driving system that follows a marker (golf balls, for example) placed in the field.

The end effector was developed using the technology of the adsorption mechanism. The conventional adsorption mechanism requires a pressing force to make it fit when sucking on the target. Agricultural products floating in the air, like a pendulum, receive almost no reaction force. Manipulators need highly accurate position control to approach crops. In this article, we describe the development of an end effector that can grip crops that can only obtain low reaction force.

Autonomous Underwater Vehicles (AUVs) have a great potential for under-ice explorations of both the Arctic and Antarctic Oceans. Survey results of current under-ice AUVs are extremely precise, but there is a trade-off between precision and running cost and/or risk of losing AUVs. Thus, the authors develop a variable and compact AUV MONACA (Mobility Oriented Nadir AntarctiC Adventurer), which can offer the necessary and sufficient compositions for various survey needs by the modular design.

This paper reports about a first autonomous navigation trial of AUV MONACA held in Oct. 2020 at Shimoda Bay, Japan. MONACA’s basic performance such as waypoint tracking and altitude/depth control were verified at the open and calm water. MONACA made 17 dives during the trial, duration was about 3 hours and travel distance was about 3 km in total. MONACA succeeded in measuring bathymetry by a multibeam sonar.

Nowadays, a variety of robots has been studied and developed for undersea exploration. It is necessary for the robots to work long time in water so as to efficiently perform given tasks. We proposed a system to make the robots be able to come back power supply unit and automatically charge battery underwater. We confirmed ROV have conducted docking that assumes charging battery under water with the system, and shown it effective. However, 3-ball marker that was used as the recognition target of the power supply unit, there were cases where two of the three balls overlapped from a certain position and could not be recognized well. Therefore, by changing from a 3-ball marker to a 4-ball marker, the remaining 3 balls can be seen even if the balls overlap, so it is easier to recognize. In this paper, we report the results of a comparison between of a 3-ball marker and a 4-ball marker on measuring accuracy of position and posture.

AUVs (autonomous underwater vehicles) for underwater research are attracting attention for practical use in the fields of research and marine development. Positioning the self-position in water is an important function in AUV. There are many methods for positioning the self-position in the air, but when it comes to positioning in the water, radio waves cannot be used, so ultrasonic waves were used. In order to create an inexpensive and high-performance underwater self-positioning system, we created a microphone circuit and speaker circuit, conducted a microphone sensitivity experiment underwater, and conducted a prediction experiment of the distance between microphones using the propagation time difference of ultrasonic waves.

In JAMSTEC, a 3D modeling technology of deep seafloor has been developed to acquire quantitative data for use in various academic surveys and discussions on resource development. SfM (Structure from Motion) technology is expected to become a powerful tool for creating 3D models of deep-sea environments where GPS and LIDAR cannot be used. Furthermore, if there are images taken at the same point in the past, it will be possible to quantitatively clarify changes before and after an event such as an earthquake. In this paper, we introduce our SfM technology and report a practical example of applying to a video acquired by Shinkai 6500.

Kobe city has been developed as port city and this effort is still regarded as important, so that marine and maritime education is also regarded as important. Therefore, training engineers is a major task for educational institution in the city. Authors proposed utilization of underwater robot in technical college of technology in 2 ways; practical class in lower grade and Project Based Leaning class in higher grade. In this article, educational underwater robot named CHVIS is developed in PBL class, with presenting theme for student and students' own voluntary efforts. This activity may have some learning effect at PBL. Learning effect is continuously verified at classwork.

In order to achieve 3D tracking of swimming organisms by an AUV (Autonomous Underwater Vehicle), we proposed a method to simultaneously estimate the position and vertical motion of sea turtles from multibeam imaging sonar images using deep learning. We trained the CNN based detector using not only the sonar images but also information about vertical motion, obtained by a data logger attached to the sea turtles. Two methods were examined: classification of the sea turtle state by the amount of depth change and regression of the pitch angle. It was also confirmed that there was a correlation between the amount of depth change and pitch angle. The effectiveness of the method was confirmed by data acquired at an aquarium.

In this research, we develop a diving support system for the purpose of supporting safety measures for the diving experience. Radio waves are highly attenuated in water, making it difficult to communicate. Therefore, communication is performed using ultrasonic waves, which have less attenuation than radio waves. In this paper, we propose an inexpensive voice transmission system and an acoustic positioning system as support systems. In the voice transmission system, voice transmission is performed between divers on the ground and underwater. In the acoustic positioning system, the position is estimated assuming that it will be mounted on a diver or an underwater drone.

The water area observation applications have been gotten much attention to identify and study the water environmental changes. We have developed a novel omni-directional movement sensing device that can move to an arbitrary position and keep its position on the water surface. The characteristics of the proposed device are portable dimension, low-cost manufacturing, omni-directional mobility, and sustainable observation. Our previous study figured out an efficient mobility strategy is necessary for the long term operation in water area. In this paper, we set up movement strategies for keeping the sensing device position in real field and compare them to identify the most efficient one.

Recent years, remotely operated vehicle (ROV) has been used for observation of a marine structure or bottom of a ship, and used for education. On the other hand, the ROV is easy to be affected by the tidal current and easy to lose its localization. So we have developed a semi-submersible autonomous vehicle that works with ROVs. Our semi-submersible type ASV has a large payload and low influence of wave, and it can be as a landmark for ROVs because of its omni-directional maneuverability. In this paper, we described the design of our semi-submersible type ASV which supports ROV's operation.

We propose a motion control method for passive towed robots focusing on turning. In the conventional method, the formation is formed by controlling the tether angle to the target value. This method assumes that the tension is maintained and does not take into account the factor of tension loss during a turn. It is considered that the target tether angle should be changed to an appropriate value in order to achieve an appropriate turn. First, we focused on the relationship between the tether angle and swing angle during a turn and derived a theoretical equation to derive the target tether angle from the target swing angle.Next, we propose a method of operating the rudder to control turning while generating brake force to apply the theoretical equation. Finally, we verified the correctness of the theoretical equation by experiments.

In this research, Haptic Shard Control (HSC) was applied to a pipe tracking task of a Remotely Operated Vehicle (ROV). HSC works with an operator to control the ROV through a single control device that can display forces to the operator’s hand. We conducted a subject experiment to compare the results of HSC and fully manual control. The experimental results show that HSC reduced Weighted WorkLoad (WWL) scores and improved the subjective evaluation of ease of operation, object recognition and so on.

Target tracking system is widely used in mobile surveillance cameras and intelligent transportation systems. Robots equipped with target tracking system work in various fields. According to the field conditions, such as in the water or in the air, the weight of the robot is limited. In addition, the radio wave attenuation of underwater robots in the water is significant. Therefore, laser is being used to replace radio wave The research of mobile communication. The master and slave of the robot are equipped with transmitter and receiver to irradiate laser to each other's receiver for communication. In the case of underwater robot, the receiver and transmitter need to follow each other's direction even when the underwater robot is moving.

Predicting fish feeding behavior is crucial in improving marine aquaculture production. We are addressing this problem using a novel approach – measuring underwater currents generated by fish, especially at feeding times. In this paper, we present our development of a suite of sensors that will collect these measurements, as we investigate their relationship with fish behavior during feeding. We added we have further developed our sensor suite by adding another current sensor for calibrating the previously developed propeller sensors and tested them along a narrow river to examine the relationship of their measurements. While a strong linear relationship was observed between the measurements, it is difficult for the propeller sensors to measure currents accurately at slower currents. Experiment at the fish farm is still needed to determine if this constraint is acceptable in our application.

Today, the introduction of robots is desired for exploration in unknown environments and investigations of disaster sites. Intelligent robot is required to work in an environment where it is constantly changing. In our laboratory, we have been studying visual servoing by stereo vision. Then, we are trying to introduce FPGA for shortening real-time measurement. FPGA stands for“Field Programmable Gate Array.” It has already been applied in a wide range of fields such as communications, medical care, and analysis. FPGA can easily change integrated circuits and perform task processing in parallel. This can be expected to speed up the calculation of goodness of fit and increase the number of GA evolutions. However, it has not been fully introduced yet. In this report, we report the realization of the calculations of a three-ball model-based matching by FPGA.

This paper discusses security development methods for the development of unmanned aircraft systems. By organizing the cases of unseen flights and comparing the standards of manned aircraft and remote control systems, we explore the procedures of security development that contribute to the development of unmanned aircraft systems. We will also define a reference architecture for unmanned aircraft systems using the Perdue Enterprise Reference Architecture, and discuss how airworthiness security and information security are integrated. As a result, for a security standard for manned aircraft such as DO-326A, the development procedure is available, but the evaluation of each component is considered appropriate to follow IEC 62443.

Even if the robot nursing care device designer takes measures intended to reduce the risk, the risk reduction effect expected by the designer cannot be obtained unless the actual device user responds as intended. In the risk assessment and risk reduction process mainly carried out by the designer, a method for comprehensive risk assessment and risk reduction was examined by sharing the residual risk, including confirmation by the user.

In services using statistical machine learning, safety quality management is important to reduce accidents and economic losses due to misjudgment and miscontrol by Artificial Intelligence (AI). However, research and development has just begun, and the management method has not yet been established. Moreover, the target of the guidelines published to date for quality management of AI-based systems are limited to supervised learning. In this paper, I develop AI safety and quality management methods for autonomous mobile robots that makes decisions and controls using deep reinforcement learning, which is not covered by the AI quality management guidelines. These methods focus on the quality characteristics and record quality assessment as evidence for analysis, understanding, and quantitative quality explanation. As a result, they enable understanding and sharing the achieved quality among stakeholders, and give concrete explanations to society. It is also useful for clarifying ordering conditions, identifying problems, and presenting added value.

It will be very useful in the workplace if everyone is able to teach robots workflow without any expertise in the conceptual design and programming stages. This paper reports on a prototype system that can detect objects with characteristics such as color, and automatically record and simulate picking operations.

Due to the increase in the number of remote workers, there is a need to develop a robotic system that can run errands outside. In this study, we use a wheeled robot indoors and a quadruped robot outdoors to realize the errand-running by coordinating the two robots.

Photographs are usually taken manually with a digital camera or a smart phone by cameraman. Therefore, there are few snapshots of all family members. We will solve this problem by using a mobile robot as a cameraman. We propose a method for taking photos using a mobile robot and selecting good photos considering human preferences. There are two subproblems to solve in developing a robot cameraman: how to position the robot and how to evaluate the quality of photos. In snapshots, it is difficult to uniquely decide the viewpoint because of the variety of the positional relationship between the subjects. Therefore, we take pictures of the subject from 360 degrees around, and then select the best pictures. In this work, we use deep learning and dataset that consider human preferences to evaluate the quality of photos.

Clearning up the tableware is a common task in home environments and restaurants. Automation of them would reduce the number of workers and compensate for many labor shortages. The purpose of this work is to propose an integrated system for a robot to manipulate the tableware onto a tray. Visual function of our system not only detects the objects on the table and determine the types of the tableware but also find the leavings on the dish. Planning function of our system calculates how to stack and align the tableware on the tray using two types of planner, symbolic planner and geometric planner. The proposed system achieves an efficient manipulation so that makes it possible to carry more dishes at a time. The experimental results proved the effectiveness of our approach.