Autonomous navigation on unknown uneven terrain needs a reliable traversability map that indicates potential navigation hazards such as slipping down from a slope, colliding with an obstacle, etc. This paper focuses on generating a real-time traversability map using a 3D LiDAR. The proposed method leverages a probabilistic inference model to update the terrain map, detect static obstacles, and remove moving objects simultaneously. A robust travelability map is created by considering both uneven terrain conditions and obstacles. Our experiments demonstrate its suitability for real-time navigation over a variety type of real-world environments.

Under sitting conditions, up-and-down head movements occur during the transition period from awakening to stage 2 non-REM sleep. This paper proposes a model that calculates head movements by giving time series of three sleep depths. The proposed model couples a head musculoskeletal model with a muscle activity model. The former is represented by an antagonistic drive system based on muscle contraction characteristics, while the latter varies the magnitude of muscle activity level according to sleep depth. We first measured variations in EEG/EMG around the neck and cervical angle during sleep onset, and based on these characteristics, we constructed two models and developed the proposed model. We then conducted numerical simulations to examine whether the proposed model can represent actual head movements and compared the simulation results with experimental results. The simulation results concluded that the proposed model can represent the experimental results and that the muscle activity waveforms generated in the intermediate stage have experimental characteristics. This model suggests that drowsiness can be estimated from head movements and is expected to be applied to estimate driver's drowsiness.

Legged locomotion, including walking and running has good adaptability for natural environment, therefore, many researchers have studied on legged robots and animals. Interaction between the robots/animals' body, their controller, and environment realizes the legged locomotion. Legged robots walk in different ways therefore we think that there are not only one kind of the interaction. In addition, researchers consider that the interaction is important to substantialize robot walking. However, there are not active discussion on it because there is not a defined scale about it. We focus on gaits that show well-ordered walking and hypothesize that we can classify the interaction in gait generation. In this research, we aim to quantify a transient state of walking. We propose a gait representation from a viewpoint of finite state machine. A binary matrix represents a gait. Its element has two states, that is swing state and stance state. The matrix comes from a gait diagram. We divide it by one step of a certain leg to discretize the transient state of walking. Furthermore, we convert the matrix into a vector to understand a gait easily. We call the vector “gait vector”. Our method enables us to distinguish any gaits in walking because walking continuously divides into “gait vectors” by our method. We can recognize emergence of steady gait immediately by numbering gait vectors. In this paper, we show an effect of our proposal by simulation with a six-legged robot walking. In addition, we propose a calculation method of similarity between gait vectors to classify a gait vector in a gait class numerically. Using an order and elements of gait vector, we can calculate the similarity between gait vectors. We show similarities of quadrupedal gaits. Our proposal may be an index of gait classification with “gait vector” because of its result.

To evaluate the work contents of a care worker, this paper proposes a novel lower limb posture estimation method, and a motion recognition method for the automation of care records. In our approach, wearable-embedded sensors consisting of inertial sensors and insole-type plantar pressure sensors were used to recognize the posture of the whole body during care tasks. The upper body posture was calculated from triaxial accelerations, and posture classification of lower limbs can be reached with characteristics extracted from plantar pressure distribution. To achieve accurate posture recognition considering unexpected posture in training phase and compound postures combining multiple postures, a new posture estimation method combining normal and complementary Gaussian mixture network (NACGMN) and posture-fitness function was developed. Using the time-series data of posture information, our proposed work estimation manner utilizing hidden semi-Markov models can recognize performed care works based on the frequency distribution of state transitions. In experiments, we measured and recognized transfer assistance works between a bed and a wheelchair in our laboratory and an actual care facility, and the results demonstrated the effectiveness of the proposed system.

This paper investigates synchronization in time-varying networks of nonlinear systems with sampled-data couplings. In particular, we consider the synchronization problem of systems with a time-varying network structure realized by switching between multiple network structures, including the completely disconnected network. Throughout this paper, we derive a synchronization condition for sampled-data network systems with such a time-varying network structure by introducing a concept of the average dwell time. Finally, we show a numerical simulation result to illustrate the validity of the derived condition.

This paper concerns a distributed estimation problem for large-scale systems with data acquired by sensor networks. Since each sensor node in a sensor network acquires the measurements at its timing, the measured data of all the sensors could be asynchronous and aperiodic sampled data. This paper develop a distributed observer for linear time-invariant (LTI) systems with asynchronously sampled data to estimate the whole state from such data. After deriving a stability condition for the estimation error dynamics, we show that the maximum allowable sampling interval and appropriate observer gain matrices, such that the estimation error converges asymptotically to zero, can be computed by solving the derived linear matrix inequality condition. The obtained observer can estimate the whole state of the systems without any clock-synchronization mechanism.

Recently, advanced driver assistance systems (ADAS), such as adaptive cruise control, lane keeping assistance and so on are widely spreading and these systems are improving comfortability and decreasing the number of accidents. The systems should be designed considering not only the usability but the acceptability from the driver because some drivers would switch it off if the system is not acceptable from the driver. Since the drivers' behavior is not simple but often varied even in a normal safety driving, there should be a kind of 'distribution' of the allowable driving behavior. This paper tries to construct the driver model which can simulate the variation in driving behavior in order to define distribution of the allowable driving behavior. Stochastic parameters and dead zones were used in the model structure taking into account the human nature. A method for evaluating the distribution of acceptable driving behavior is proposed, and parameter study is done based on the method.