Wireless sensor and actuator networks (WSANs) are required to achieve both energy-efficiency and low-latency in order to prolong the network lifetime while being able to quickly respond to intermittently-transmitted control commands. These two requirements are in general in a relationship of trade-off when each node operates with well-known duty-cycling modes: nodes need to make their radio interfaces (IFs) frequently active in order to promptly detect the communication requests from the other nodes. One approach to break this inherent trade-off, which has been actively studied in recent literature, is the introduction of wake-up receiver that is installed into each node and used only for detecting the communication requests. The radio IF in each node is woken up only when needed through a wake-up message received by the wake-up receiver. While the effectiveness of this type of on-demand WSANs has been shown in several studies by theoretical analysis and computer simulations, its implementation and large-scale experimental investigations are missing. Therefore, in this paper, we first design and implement radio-on-demand sensor and actuator networks (ROD-SAN) including all protocols to realize on-demand WSANs, from the lowest layer of wake-up signaling to the application layer offering the functionalities of information monitoring and networked control. Then, we show experimental results obtained through our field trial in which 20 nodes are deployed in an outdoor area with the scale of 450m X 200m. The numerical results provide us with practical insights on the effectiveness as well as limitations of on-demand WSANs.

Energy harvesting wireless sensor network (EHWSNs) are being actively studied in order to solve the problems faced by battery-operated WSNs, namely the cost for battery replacement and impacts on the environment. In EH-WSNs, each node harvests ambient energy, such as light, heat, vibration, and uses it for sensing, computations, and wireless communications, where the amount of harvested energy of each node varies depending on their surrounding environments. MAC protocols for EH-WSNs need to be designed to achieve high throughput and fairness, however, the conventional MAC protocols proposed for EH-WSNs do not adapt to the harvesting rate of each node, resulting in poor fairness. In this paper, we propose a fair MAC protocol based on polling scheme for EH-WSNs. The proposed scheme adjusts contention probability of each node according to the individual harvesting rate, thereby increasing throughput achieved by nodes with low harvesting rate. We evaluate throughput and fairness of the proposed fair polling scheme by computer simulations, and show that the proposed scheme can improve fairness with little degradation of the overall network throughput.

In order to reduce energy wastefully consumed by access points (APs) in wireless LAN (WLAN), Radio-On-Demand (ROD) WLAN has been proposed, which offers wireless access only when and where it is needed in an on-demand manner. Each AP in ROD is equipped with a wake-up receiver, which transits a state of AP from sleep to active by receiving a wake-up signal from a station (STA). However, if there are several APs having the same wake-up ID around a STA, a wake-up request from STA causes more than one AP to wake up though STA only needs a single AP. In this paper, in order to reduce the number of APs redundantly woken up in such a situation, we propose a probability-based wake-up mechanism. With the proposed mechanism, each AP does not always wake up after receiving the wake-up request. Instead, each AP decides whether it should wake up or not based on the wake-up probability. This probability is adjusted in such a way that only APs suited for serving the corresponding STA have the higher probability to wake up while the other APs maintain a lower probability. By computer simulations, we show that our proposed mechanism can significantly reduce the number of redundant wake-up while keeping the adverse impact on their usability and quality of service (QoS) small.

The MMSOM identification method, which had been presented by the authors, is improved to the multiple modeling by the irregular self-organizing map (MMISOM) using the irregular SOM (ISOM). Inputs to the neural networks are parameters of the instantaneous model computed adaptively at every instant. The neural network learns these models. The reference vectors of its output nodes are estimation of the parameters of the local models. At every instant, the model with closest output to the plant output is selected as the model of the plant. ISOM used in this paper is a graph of all the nodes and some of the weighted links between them to make a minimum spanning tree graph. It is shown in this paper that it is possible to add new models if the number of models is initially less than the appropriate one. The MMISOM shows more flexibility to cover the linear model space of the plant when the space is concave.

The MMSOM identification method, which had been presented by the authors, is improved to the multiple modeling by the irregular self-organizing map (MMISOM) using the irregular SOM (ISOM). Inputs to the neural networks are parameters of the instantaneous model computed adaptively at every instant. The neural network learns these models. The reference vectors of its output nodes are estimation of the parameters of the local models. At every instant, the model with closest output to the plant output is selected as the model of the plant. ISOM used in this paper is a graph of all the nodes and some of the weighted links between them to make a minimum spanning tree graph. It is shown in this paper that it is possible to add new models if the number of models is initially less than the appropriate one. The MMISOM shows more flexibility to cover the linear model space of the plant when the space is concave.

In this paper, a novel analysis technique is applied for investigating human operators' trial and error learning process to control a nonholonomic system, 2-link planer underactuated manipulator (2PUAM). An essential core of the technique is to use a value function of the reinforcement learning scheme for revealing how the operators can find a control strategy. It is an advantage of the proposed technique compared to the others that a transition of the value function may explain the changes of the operators' strategies during the learning process. According to the results of the analysis, the operators tended to explore an objective trajectory first, and then shift to the tracking control of the trajectory. The tracking was accompanied with acceleration to achieve the goal faster. Interestingly, the acceleration disturbs the objective trajectory due to the complex dynamics of the target, and induces another exploration to get better trajectories. The fact that this phase transition structure under unsupervised learning environment is consistent with previously reported results for a supervised case implies that the structure can be a general nature of human learning process.

Recent research on hidden-state reinforcement learning (RL) problems has been concentrated in overcoming partial observability by using memory to estimate states. Switching Q-learning (SQ-learning) is a novel memoryless approach for RL in partially observable environments. The basic idea of SQ-learning is that "non-Markovian" tasks can be automatically decomposed into subtasks solvable by memoryless policies, without any other information leading to "good" subgoals. To deal with such decomposition, SQ-learning employs ordered sequences of Q-modules in which each module discovers a local control policy. Furthermore, a hierarchical structure learning automaton is used which finds appropriate subgoal sequences. We apply SQ-learning to three partially observable maze problems. The results of extensive simulations demonstrate that SQ-learning has the ability to quickly learn optimal or near-optimal policies without huge computational burden.

A method for building time series prediction model using genetic programming is proposed. The construction of prediction models consists of two stages. The first stage composes the most appropriate functional form with genetic programming. The second stage fixes optimal parameters involved in the composite function with a backpropagation-like algorithm. The second stage can be recognized as a local search, which is a powerful tool to accelerate the evolving speed of GP and GA. The method is applied to typical time series and some real world prediction problems. Results of computer generated chaotic time series were compared to those of neural network based and autoregressive predictions. The superiority of the proposed method is demonstrated in the results of these experiments.

This paper proposes a multi-modal neural network model composed of a pre-processing module and a post-processing module in order to enhance the nonlinear characteristics of neural networks. The pre-processing module is made of Chebyschev polynomials and transforms input data into spectra. The post-processing module is made of multilayer neural network and associates according to the spectral inputs generated by the pre-processing module. Fundamental experiments upon pattern recognition and functional approximation and experiments applying the method to a control problem result that the method enable one to build small scale neural model for nonlinear systems and to perform learning in shorter time.

We consider the problem of recognizing the driving environment of a vehicle by using information obtained from some sensors of the vehicle. Previously, we presented a recognition algorithm based on an usual method of pattern matching by use of fuzzy reasoning. Furthermore, this algorithm was extended to meet the demands of nonstandard drivers and changes of vehicle properties. In this algorithm we supposed that an extra source of knowledge (supervisor) for correcting the decision taken by the classifier could be acquired, and inevitably we expected additional sensors. To cover such weakness we present a self-supervised learning algorithm. Here the supervisor is constructed by using a-cut of the membership function of the representative class. Computer simulation on practical uses of the vehicle shows efficiency of the self-supervised learning.

A simple discrete-time design of robust iterative learning controllers taking account of the transient behavior as well as the uncertainty of a plant is proposed. Using the impulse response sequence of a plant, a simple finite-dimensional formulation of the problem is given. Assuming that a nominal impulse response sequence is given, a design based on the minimization of a quadratic performance index that can be regarded as a measure for the transient performance is proposed. The effect of the error in the impulse response data is analyzed. It is shown that an excessively high-order controller is not robust in the sense that the error severely deteriorates the transient performance. To obtain a robust controller with a reasonable order, a design based on a probabilistic modeling of the error in the impulse response data is proposed. The controller is obtained by minimizing an averaged quadratic performance index. Simulation examples are presented to illustrate the effectiveness of the proposed methods.>

The use of a robot to assemble many parts on a board is addressed. When there exists a large variety of part types, it is an important task to find a sequence by which the parts can be assembled in a short time. This problem can be regarded as an extension of the efficient drawing path problem. The stepwise clustering method, which has been very effective for the latter problem, is extended to provide an efficient assembly sequence and is applied to the assembly sequence of electronic circuit boards. The resulting sequences are shown to be efficient enough to give shorter assembly times than those by human experts.>

An efficient scheme is presented for a learning control problem of finite Markov chains with unknown dynamics, i.e. with unknown transition probabilities. The scheme is designed to optimize the asymptotic system performance and for easy application to models with relatively many states and decisions. In this scheme a control policy is determined each time through maximization of a simple performance criterion that explicitly incorporates a tradeoff between estimation of the unknown probabilities and control of the system. The policy determination can be easily performed even in the case of large-size models, since the maximizing operation can be greatly simplified by use of the policy-iteration method. It is proven that this scheme becomes epsilon -optimal as well as optimal by suitable choice of control parameter values in the sense that a relative frequency coefficient of making optimal decisions tends to the maximum.>

This paper presents a simple and effective algorithm for an adaptive multifrequency receiver that can be used for several types of multifrequency signals. This algorithm uses the ability of frequency detection of signal processing using extreme and zero-crossing points in the multifrequency signals and an adaptive second-order notch filter.

A simple algorithm for the finite-time two-armed bandit problem is proposed. In this algorithm, the whole process is divided into the first estimating process and the next controlling process. Efficient methods involving the use of approximation for computing the optimal length of the estimating process are provided.