並木 明夫

With the development of science and technology, the traditional industrial structures are constantly being upgraded. As far as drones are concerned, an increasing number of researchers are using reinforcement learning or deep learning to make drones more intelligent. At present, there are many algorithms for object detection. Although many models have a high accuracy of detection, these models have many parameters and high complexity, making them unable to perform real-time detection. Therefore, it is particularly important to design a lightweight object detection algorithm that is able to meet the needs of real-time detection using UAVs. In response to the above problems, this paper establishes a dataset of six animals in grassland from different angles and during different time periods on the basis of the remote sensing images of drones. In addition, on the basis of the Yolov5s network model, a lightweight object detector is designed. First, Squeeze-and-Excitation Networks are introduced to improve the expressiveness of the network model. Secondly, the convolutional layer of branch 2 in the BottleNeckCSP structure is deleted, and 3/4 of its input channels are directly merged with the results of branch 1 processing, which reduces the number of model parameters. Next, in the SPP module of the network model, a 3 x 3 maximum pooling layer is added to improve the receptive field of the model. Finally, the trained model is applied to NVIDIA-TX2 processor for real-time object detection. After testing, the optimized YOLOv5 grassland animal detection model was able to effectively identify six different forms of grassland animal. Compared with the YOLOv3, EfficientDet-D0, YOLOv4 and YOLOv5s network models, the mAP_0.5 value was improved by 0.186, 0.03, 0.007 and 0.011, respectively, and the mAP_0.5:0.95 value was improved by 0.216, 0.066, 0.034 and 0.051, respectively, with an average detection speed of 26 fps. The experimental results show that the grassland animal detection model based on the YOLOv5 network has high detection accuracy, good robustness, and faster calculation speed in different time periods and at different viewing angles.

<title>Abstract</title> Operation of tools has long been studied in robotics. Although appropriate hold of the tool by robots is the base of successful tool operation, it is not with ease especially for tools with complicated shape. In this paper, an assist system for a four-limbed robot is proposed for remote operation of reaching and grasping electric drills using deep reinforcement learning. Through comparative evaluation experiments, the increase of success rate for reaching and grasping is verified and the decrease in both physical and mental workload of the operator is also validated by the index of NASA-TLX.

In recent years, tracking of articulated objects at high speed with monocular cameras has been gaining attention. This study presents a novel method for high-frame-rate articulated object tracking with a monocular camera. The method is an extended version of our previous research on high-speed monocular rigid object tracking. In this study, to realize tracking of an articulated object, we integrate dual-quaternion kinematics with our previous fast pixel-wise-posteriors (fast-PWP3D) tracking framework, and propose an auto-regressive (AR) process to encode the dynamic propagation of the estimated state vectors. We give a full three-dimensional derivation of the mathematical formulation of our method and show that our method is capable of tracking an articulated object having a large number of degrees of freedom with only a monocular camera, and is robust against dynamic environmental changes (e.g., illumination/partial occlusion). Moreover, we show an efficient implementation strategy of our method. The results of real-time experiments show that we achieved nearly 350 Hz performance when tracking a four degrees-of-freedom (4-DOF) articulated object with a monocular camera. &copy; 2020 IEEE.

In this paper, we propose the "multi-vision hand", in which a number of small high-speed cameras are mounted on the robot hand of a common 7 degrees-of-freedom robot. Also, we propose visual-servoing control by using a multi-vision system that combines the multi-vision hand and external fixed high-speed cameras. The target task was ball catching motion, which requires high-speed operation. In the proposed catching control, the catch position of the ball, which is estimated by the external fixed high-speed cameras, is corrected by the multivision hand in real-time. In experiments, the catch operation was successfully implemented by correcting the catch position, thus confirming the effectiveness of the multi-vision hand system.

In recent years, high-speed vision systems are increasingly used for robot control. However, robot systems developed to date have not sufficiently utilized rapid responses of high-speed vision due to poor performance of control methods. In order to realize high-speed robot operation, it is necessary to design a control method to use the maximum torque generated by the robot. In this paper, we propose a control method integrating high speed visual processing and real time optimum control. This control method is based on nonlinear model predictive control (NMPC), and optimizes the response of the position of the robot arm so as to accurately follow the target trajectory. In this paper, we describe system design to integrate high speed visual processing and NMPC, taking high-speed operation of air hockey robot as an example. Finally, we present experimental data and verify the effectiveness of the system.

This paper demonstrates the relationship between the production process of a knot and manipulation skills. First, we define the description (rope intersections, grasp type and fixation positions) of a knot. Second, we clarify the characteristics of the manipulation skills from the viewpoint of the knot description. Next, in order to obtain the production process of the knot, we propose an analysis method based on the structure of the knot and the characteristics of the manipulation skills. Using the proposed analysis method, we analyzed eight kinds of knots, formed with a single rope, two ropes or a single rope and an object. Finally, in order to validate the production process obtained by the proposed analysis method, we show experimental results of an overhand knot and a half hitch produced by using a robot hand system.

In this paper, we propose a method for improving the maneuverability of master-slave systems. We aim at reproducing human skillfulness and dynamic performance in master-slave robots by using assist control for human operators. In this paper, we tackle a reaching task performed by a master-slave robot and propose an operation assist algorithm based on visual feedback control. The algorithm consists of visual object recognition for a slave robot, prediction of the operator's motion by a particle filter, estimation of the operator's intention, and operation assistance of the reaching motion. We verified the validity of the proposed system in experiments.

Mobile robotics is a potential solution to home behavior monitoring for the elderly. For a mobile robot in the real world, there are several types of uncertainties for its perceptions, such as the ambiguity between a target object and the surrounding objects and occlusions by furniture. The problem could be more serious for a home behavior-monitoring system, which aims to accurately recognize the activity of a target person, in spite of these uncertainties. It detects irregularities and categorizes situations requiring further explorations, which strategically maximize the information needed for activity recognition while minimizing the costs. Two schemes of active sensing, based on two irregularity detections, namely, heuristic-based and template-matching-based irregularity detections, were implemented and examined for body contour-based activity recognition. Their time cost and accuracy in activity recognition were evaluated through experiments in both a controlled scenario and a home living scenario. Experiment results showed that the categorized further explorations guided the robot system to sense the target person actively. As a result, with the proposed approach, the robot system has achieved higher accuracy of activity recognition.

In developing a shoulder prosthesis, in addition to appropriate payload and range of motion under the constraints of weight and shape, impact absorption is very important for safe use. Hybridization of two different actuators (pneumatic elastic actuators with the features of lightness and intrinsic visco-elasticity, and servo motors that have stable torque and a large range of motion in combination with an antagonistic mechanism) was employed to achieve the development of the shoulder prosthesis. A two-link, two-degree-of-freedom arm was used to test the different hybridization configurations in order to investigate the impact absorption. A dynamic simulation platform based on four bimanual activities of daily living was established to obtain the required range of motion and torque for joints of a two-link, four-degree-of-freedom arm. The number of pneumatic elastic actuators required and the dimension of the antagonistic mechanism mechanical structures were optimized using the dynamic simulation platform. The best configuration of the two types of actuators was determined using the dynamic simulation based on the impact absorption results and other criteria. Moreover, a simplified prototype driven by hybrid actuation was made. It was shown that the pneumatic elastic actuator joint could improve impact absorption, and the actuator configuration of shoulder prostheses is activity of daily living dependent. The prototype could reproduce a certain activity of daily living motion, indicating its feasibility in daily living.

In this paper, we propose an entirely new manipulation strategy for dynamic manipulation of a rhythmic gymnastics ribbon, as one example of belt-like flexible objects, with a high-speed multi jointed manipulator. The manipulation strategy involves manipulating the target object (rhythmic gymnastics ribbon) at a constant, high speed. Then, we can assume that the dynamic behavior of the ribbon can be obtained by performing algebraic calculations of the robot motion using the proposed strategy. Based on this assumption, we derive a simplified deformation model of the ribbon and suggest a simple motion planning method using the proposed model. Finally, we show simulation results and experimental results of shape generations (for example, circle, wave, figure-eight, and crack shapes) of a ribbon based on the proposed method, and we discuss quantitative evaluation of shape generations by image processing.

Multifingered robot hands have the ability to manipulate and grasp various objects, like a human hand. In previous studies, we have looked at paper folding operations (Origami) as an example of dexterous manipulation to be performed by dual robot hands. It is difficult for robot hands to manipulate a sheet of paper because of its deformation, and it is thus necessary to recognize its shape during Origami operations. In this paper, we propose a method of estimating a 3D model of paper based on depth information obtained from a 3D sensor. The 3D model of the paper is expressed by using a physics simulator composed of nodes with springs and dampers. As a result, highly accurate 3D shape estimation was achieved during Origami operations.

In recent years, the demand for robots that can perform various tasks in dangerous environments has increased. Teleoperated robots are more suitable for dangerous environments than autonomous robots. We have developed a master-slave robot system that consists of a lightweight master device and a high-power slave robot and proposed operation assistance methods to improve the dexterity and increase the moving speed. In this paper, we propose a control method for catching a flying ball with a master-slave system. The slave robot is controlled by integrating the operator's commands and autonomous visual feedback control so as to track and catch the flying object. The object trajectory and operator motion are predicted by using a Kalman filter. The proposed method was verified by experiment.

In this study, we propose a novel algorithm for an air-hockey robot to select an optimal attack motion according to the behavior of a human opponent. The attack motion of the robot is optimized during an air hockey game by two-step look ahead prediction. First, we explain how to calculate state values in continuous games like air-hockey. Second, we model the human and robot behaviors and propose a value function to be used in the optimization. Then, we show experimental data of simulations and analysis results of the robot behaviors generated from the calculated value.

High-speed recognition of the shape of a target object is indispensable for robot arms to perform various kinds of dexterous tasks in real time. In this paper, we propose a high-speed 3-D sensing system with active target-tracking. The system consists of a high-speed camera head and a highspeed projector, which are mounted on a two-axis active vision system. By measuring coded structured light projected by the projector, 3-D measurement at a rate of 500 fps is achieved. The measurement range was increased because of the active tracking, and the shape of the target was accurately observed even when it moved quickly.

ln the paper, we propose a method for improving maneuverability of master-slave systems. We aim for reproducing human skillfulness and dynamic performance in master slave robots by using assist control for human operators. In this paper, we focus on a reaching task of a master-slave robot and propose an operation assist algorithm based on visual feedback control. It consists of visual recognition of an object for a slave robot, prediction of operator's motion by a particle filter, estimation of a target for grasping, assist control of the reaching motion. Finally, the validity of the proposed method is verified in a master-slave robot system.