Minoru Sekiguchi

Mobile robot control by a structured hierarchical neural network

A mobile robot whose behavior is controlled by a structured hierarchical neural network and its learning algorithm is presented. The robot has four wheels and moves about freely with two motors. Twelve sensors are used to monitor internal conditions and environmental changes. These sensor signals are presented to the input layer of the network, and the output is used as motor control signals. The network model is divided into two subnetworks connected to each other by short-term memory units used to process time-dependent data. A robot can be taught behaviors by changing the patterns presented to it. For example, a group of robots were taught to play a cops-and-robbers game. Through training, the robots learned behaviors such as capture and escape.<<ETX>>

Dialog interface system

In the dialog interface apparatus of the present invention, input speech is converted to an input semantic representation by a speech recognition unit, and a dialog management unit outputs an output semantic representation that corresponds to the input semantic representation, based on the input semantic representation obtained by the speech recognition unit. Having received the output semantic representation from the dialog management unit, a speech synthesis unit converts the output semantic representation to output speech identifying a specific dialog target and outputs the output speech. Further, the dialog management unit outputs to an innate operation execution unit an innate operation command that corresponds to the input semantic representation. The innate operation execution unit receives the innate operation command from the dialog management unit and executes an operation corresponding to the innate operation command.

Mobile robot control by neural networks using self-supervised learning

A reinforcement learning algorithm based on supervised learning is described. It uses associative search to discover and learn actions that make the system perform a desired task. One problem with associative search is that the systems actions are often inconsistent. In the searching process, the systems actions are always decided stochastically, so the system cannot perform learned actions more than once, even if they have been determined to be suitable actions for the desired task. To solve this problem, a neural network that can predict an evaluation of an action and control the influence of the stochastic element is used. Results from computer simulations using the algorithms to control a mobile robot are described. >

Exploring Everyday Activities for Pervasive Decision-Aid

In this paper, we explain the process of establishing shoppers activity-models based on a series of user-researches as the basis of pervasive systems for a supermarket. Pervasive systems have been recognized as the technologies which enable users decision-aid in his/her everyday-activities. For instance, a smart travel navigation system, which employs embedded and wearable devices and mobile agent technologies, was proposed as a promising information system for the society: it renders complex tasks into simple subtasks including providing adequate information regarding transit to the other train at an arbitrary station for visually-impaired person. Few researches on analysis of users everyday-activities for the systems design, however, have been conducted. We believe important to examine users everyday-activities as well as to develop elemental technologies of pervasive systems simultaneously, which will become a powerful way of solving a variety of real-world problems. An important knowledge regarding the model is that a shopper gradually elaborates vague plans primarily conceived at off the store into final decision-making at checkouts, instead of buying items according to well-defined plan as well as just on impulse. We regarded that the system dynamically adapting to these shoppers contexts is very different from other shoppers decision-aid systems

Behaviour control for a mobile robot by a structured neural network

We have been researching ways to use neurocomputers that have highly parallel data processing and learning functions for robot control. In this paper, a structured network model for robot control and its learning algorithm are presented. There are three requirements for the robots: (1) The robot must be easy to control but the neural network must be sophisticated enough to handle multiple sensor input. (2) The robot must be able to learn easily. (3) The robot must be able to adjust its own actions. We have developed a new mobile mechanism, created a network model, and increased the network learning speed. Sensor signals from the robot are input to the neural network. The network outputs a certain reaction pattern in response to the sensor input. Then the reaction is refined to an ideal one using training patterns. A robot can change its reaction pattern by changing the training pattern. We created two robots with different action patterns: one chases other robots and the other runs away from other robots....

Neural network-based landmark detection for mobile robot

The mobile robot can essentially have only the relative position data for the real world. However, there are many cases that the robot has to know where it is located. In those cases, the useful method is to detect landmarks in the real world and adjust its position using detected landmarks. In this point of view, it is essential to develop a mobile robot that can accomplish the path plan successfully using natural or artificial landmarks. However, artificial landmarks are often difficult to construct and natural landmarks are very complicated to detect. In this paper, the method of acquiring landmarks by using the sensor data from the mobile robot necessary for planning the path is described. The landmark we discuss here is the natural one and is composed of the compression of sensor data from the robot. The sensor data is compressed and memorized by using five layered neural network that is called a sand glass model. The input and output data that neural network should learn is the sensor data of the robot that are exactly the same. Using the intermediate output data of the network, a compressed data is obtained, which expresses a landmark data. If the sensor data is ambiguous or enormous, it is easy to detect the landmark because the data is compressed and classified by the neural network. Using the backward three layers, the compressed landmark data is expanded to original data at some level. The studied neural network categorizes the detected sensor data to the known landmark.

Parallel-distributed mobile robot simulator

The aim of this project is to achieve an autonomous learning and growth function based on active interaction with the real world. It should also be able to autonomically acquire knowledge about the context in which jobs take place, and how the jobs are executed. This article describes a parallel distributed movable robot system simulator with an autonomous learning and growth function. The autonomous learning and growth function which we are proposing is characterized by its ability to learn and grow through interaction with the real world. When the movable robot interacts with the real world, the system compares the virtual environment simulation with the interaction result in the real world. The system then improves the virtual environment to match the real-world result more closely. This the system learns and grows. It is very important that such a simulation is time- realistic. The parallel distributed movable robot simulator was developed to simulate the space of a movable robot system with an autonomous learning and growth function. The simulator constructs a virtual space faithful to the real world and also integrates the interfaces between the user, the actual movable robot and the virtual movable robot. Using an ultrafast CG (computer graphics) system (FUJITSU AG series), time-realistic 3D CG is displayed.

Application of neural networks for self-supervised learning

The learning method of layered neural networks can be supervised or unsupervised. Back propagation learning algorithm is a common method of supervised learning that can learn automatically from teaching patterns. However, accurate teaching patterns are not always available for robotic applications and it is necessary to devise a method of producing them. In this paper, two applications of neural network for self-supervised learning are described. One is a system for which a mobile robot learns its behavior by automatically generating and self- evaluating teaching data through a random walk. The other is a control method of an inverted pendulum using a knowledge-based neural network. The system collects the state data of the inverted pendulum such as angles and angular velocities by trial and error. After that, the system generates teaching data by comparing the collected data with stored knowledge. This knowledge expresses the ideal status of the inverted pendulum when it inverts. The system learns from the generated teaching data and the pendulum inverts stably after some trial and error. In both systems, the neural network learns the teaching data that is generated by the system itself.

Control of an inverted pendulum by a neural network with self-supervised learning

Summary form only given, as follows. The authors propose an adaptive self-supervised learning system based on a neural network with supervised learning. The adaptive system learns the desired task autonomously. Although this system, like many adaptive learning systems, uses trial and error, experience rules are implemented into the system as an equation so that the system can effectively generate training data based on the experience rules during trial and error and train the neural network controlling the system itself via supervised learning. The authors discuss control of an inverted pendulum to show how the adaptive system is used. The system was able to invert the pendulum stably at the target position.<<ETX>>