Keita Nakamura

Visualization of similar news articles with network analysis and text mining

This paper proposes the method to classify news articles by combining tf-idf and n-gram. This method extracts characteristic words from each news article and classify news based on these words. Numerical experiment results show the relationship among the news articles and visualize the similar articles with networks. Additionally, the authors compare proposal method with only tf-idf in order to verify the effectiveness of this method.

Quantitative Evaluation of Streaming Image Quality for Robot Teleoperations

To develop a measure of streaming video quality for remotely operated robots, we need to know the critical factors for the quality of control. Controlling robots remotely is crucial for disaster response, and many attempts have been made to create such systems. Wireless communication, which is used in remote-control systems for unmanned vehicles, changes dynamically and the streaming quality also changes with the quality of the network. However, wireless conditions are not typically measured in conventional robot systems. In this paper, to develop a quality measure for remote control using video properties, we investigate critical factors such as delay and the degradation of image quality. We also introduce the concept of a quality-of-control measure using delay and degradation of image-quality curves from simulation environments, and we discuss the impacts of changing communication delay and degrading image quality on remote-control robots.

Robot sweep path planning with weak field constrains under large motion disturbance

The sweeping robot plans a path and moves along its prior path. In conventional studies, the target field is separated into square cells to enable the robot to sweep evenly. The prior sweep path is generated by passing all the target cells. However, an outdoor sweeping robot cannot move as expected, because the robot cannot go to the next target easily due to the uncertainty of the motion of the robot. The uncertain motion is caused by individual differences of motors, disturbances from the environment, and position error. As a result, the robot passes the same point many times and the actual path length becomes longer. In this study, we propose sweep path planning to solve this problem by decreasing the number of cells that the robot must pass. Numerical simulations are carried out to verify our method and to verify the relation among the sweeping rate and robot disturbances. Simulation results show that our method is effective enabling the robot to satisfy a sweep rate of 80% and more, even if the robot has uncertainty of movement.

Stable pulling out motion for a dual-arm robot

Many disaster response robots have been studied and developed to reduce the risk of secondary disaster. These robots are expected to improve efficiency and safety. In this paper, we consider a task that a dual-arm disaster response robot pulls a bar whose length and mass are unknown out of a wall as debris removal. Safe working is important in the disaster site to remove debris. Therefore, our objective is to develop a stable pulling out motion. To achieve pulling a target bar out stably, we need to know the physical parameter of the bar. Therefore, the robot uses force/torque sensors attached to the wrists of the robot to estimate the mass and center of gravity position of the bar. Then, the robot controls the bar attitude with the estimated parameters after the bar is pulled out of the wall. Experimental results for verification show the effectiveness of the proposed motion.

Mechanism of generating drawbar pull of rod wheel on loose soil

In this paper, we propose a model of drawbar pull generated by wheels fitted with a rod and assess it by comparing measured values obtained from an experiment with those from the model. In recent years, many kinds of robots for weeding in paddy fields have been developed. However, almost all of these are large and heavy. We have previously developed a small, lightweight robot for weeding. This robot is equipped with a rod wheel that has roles of weeding and running. However, this wheel was developed by experience from demonstrations and its dynamics for control remain unknown. To solve this problem, we propose a new model for drawbar pull generated by rod wheels and evaluate it by comparing experimental values with those from the model.

Collision identification in weeding robot with acceleration standard deviation

Recently, in the background such as the organic agriculture has been focused for healthy life, a weeding robot for a rice field, called “Aigamo robot”, has developed in our research group. It stirs soil by moving on the rice field, and prevent weed seedling from establishing. For controlling it in autonomously, it needs to identify if it is at an edge on the field and makes a turn. However, GPS (Global Positioning System) module does not give us enough accuracy for the above identification. On the other hand, ultrasound sensors cannot use because the robot works in muddy soil which absorbs ultrasonic. Therefore, in this research, we develop a collision identification method by identifying if the robot is in motion or not with the acceleration sensor, and verify it in an actual rice field.

Formulation and solution method of tour conducting knapsack problem

Many studies for developing methods of solving have been done in the field of combinatorial optimization. However, it cannot simply apply these methods to real problems. For example, when a tourist decide to traveling schedule within a traveling time limit, he or she needs to select and travel tourist spots in order to be satisfied as far as possible. Such a problem cannot be solved by conventional solving method. In this study, the authors define this problem as TCKP (Tour Conducting Knapsack Problem). And the authors formulate and develop method for solving TCKP. Formulation and Solution method of TCKP are based on those of TSP (Traveling Salesman Problem) and knapsack problem. Numerical experiment is carried out in order to verify the effectiveness of proposed method. Experimental results show that it can be obtained the optimized solution within one minute when the number of tourist spots is 15 and fewer.

Emergence of adaptive behaviors for artificial creature using a combined artificial neural network

This study proposes a method to acquire adaptive behavior for artificial creature which has a lot of joints using a combined Artificial Neural Network (ANN). Experiment in this study focuses on artificial fish model, which has a lot of joints, tracking towards a target in the virtual water environment. In order to control motions of joints, a combined ANN is implemented with the model. At first, one ANN is prepared to control specific joints so as to swim basically in response to minimal input information using evolutionary computation in preliminary experiments. And an new network is constructed by combining its network and the other network. In order to acquire complicate behavior for artificial creature, weights of combined ANN are optimized. Experiment result shows the model which has many joints acquire adaptive swimming behavior towards a target by optimizing combined network.