Automatic speech recognition (ASR) is essential for human-humanoid communication. One of the main problems with ASR by a humanoid is that it is inevitably generates motor noises. These noises are easily captured by the humanoid's microphones because the noise sources are closer to the microphones than the target speech source. Thus, the signal-to-noise ratio (SNR) of input speech becomes quite low (sometimes less than 0 [dB] ) . However, it is possible to estimate these noises by using information on the humanoid's motions and gestures. This paper proposes a method to improve ASR for a humanoid with motor noises by utilizing its motion/gesture information. The method consists of noise suppression and missing-feature-theory-based ASR (MFT-ASR) . The proposed noise suppression technique is based on spectral subtraction, and a white noise is added to blur distortion of suppression. MFT-ASR improves ASR by masking unreliable acoustic features in the input sound. The motion/gesture information is used for obtaining the unreliable acoustic features. Furthermore, we also evaluated with the acoustic model adaptation technique called MLLR (Maximum Likelihood Linear Regression) . Un-supervised MLLR was used for the adaptation. We evaluated the proposed method through recognition of speech recorded by using Honda ASIMO in a room with reverberation. The noise data contained 34 kinds of noises: motor noises without motions, gesture noises, walking noises, and other kind of noises. The experimental results show that the proposed method outperforms the conventional multi-condition training technique.
We developed a smart handcart with autonomous returning function. In the conventional handcarts to carry baggage, a user pushes a handcart with baggages on the way to destination, and then, he also should carry it back without load. The developed handcart is equipped with odometry function and a laser range scanner. It records its trajectory and surrounding environment while going to destination, and it can autonomously return back to the station by retracing the recorded trajectory and environmental information. As feature of the environment to compensate the accumulating error in odometry, the cart uses flat wall and edge of pillar which is extracted by its laser range scanner. Experimental results shows the effectiveness and reliability of the system.
This paper introduces an active vision system using a parallel mechanism. Basically, active vision systems are kinds of intelligent machines using vision sensors and they track targets on screens. They are utilized as shape recognotion apparatuses, shape inspection systems, and vision systems of mobile robots, for example. Most of active vision have adopted serial link manipulators so far. They have ability to move in wide space but they need large space when they change thier orientation. Meanwhile, the active visions using parallel mechanisms have merits that they can change their orientations with small motions and position errors of active vision are averagely distributed to thier all links. Furthermore, the active vision systems have redundant degrees of freedoms relative to the degrees of freedoms on screens. This paper presents a control technique making use of the redundancy. In particular, the presented control system performs to track a target point in three cases of basic control, translation control, and orientation control. And finally, the experimental results are shown in order to evaluate the prestnted technique.
Previously, the authors have proposed a scheme for collective control of a swarm of autonomous mobile robots. It enables a human operator to effectively control the geometrical center of the swarm only via bird-view observation and a broadcast command. However, for such control, it seems necessary that the operator and the every agent of the swarm refers to the same coordinate system. In this paper, first, the influences of deviations of coordinates among the operator and the agents due to dead-reckoning are examined theoretically. The result shows that, though the deviation between the operator and the average of the agents can easily be compensated by the operator, those among the agents lead to poor maneuverability and performance of the collective manual control. Thus, a method called COMPASS-CC is introduced which can merge the coordinate systems close enough among all the agents in a fully distributed way and without any global localization scheme, by extending the previously proposed method COMPASS. Besides, its effectiveness is analyzed both theoretically and by simulation. The equation for the practical estimation of the steady-state performance is derived and verified. Finally, the validity of the method is confirmed through manual control experiments using a swarm control simulator.
This paper describes a view planning of multiple cameras for tracking multiple persons. View planning of cameras is a very important problem in watching multiple persons in a wide area by using a few cameras. We select fixation points of cameras so that the expected number of tracked persons is maxmized, based on a probabilistic model of person motion. We propose a multi-start local search-based method for tracking persons intermittently using a criterion which allows frequent shifts of fixation points. This view planning outperforms the others and is considered to be appropriate for wide area surveillance systems. We then modify the method so that the planning cost is reduced. We divide the cameras into mutually independent groups based on relations between their veiwing directions and determine fixation points within each group. The performance of this modified method is comparable to the original one with a lower planning cost.
In order to construct three-dimensional shape models of large-scale architectural structures using a laser range finder, a number of range images are normally taken from various viewpoints, and these images are aligned using post-processing procedures such as the ICP algorithm. However, in general, before applying the ICP algorithm, these range images must be registered to roughly correct positions by a human operator in order to converge to precise positions. In addition, range images must be made to sufficiently overlap each other by taking dense images from close viewpoints. On the other hand, if the positions of the laser range finder at viewpoints can be identified pre-cisely, local range images can be directly converted to the global coordinate system through a simple transformation calculation. The present paper proposes a new measurement system for large-scale architectural structures using a group of multiple robots and an on-board laser range finder. Each measurement position is identified by a highly precise positioning technique called Cooperative Positioning System (CPS), which utilizes the characteristics of the multiple-robot system. The proposed system can construct 3D shapes of large-scale architectural structures without any post-processing procedure such as the ICP algorithm or dense range measurements. Measurement experiments in unknown and large indoor/outdoor environments are successfully carried out using the newly developed measurement system consisting of three mobile robots named CPS-V.
In this paper, we propose a method to estimate a human's positions and postures by utilizing tactile information collected by a robot. We build a map that describes the correspondences of tactile information to the human's positions and postures. This map consists of clusters of tactile sensor data patterns and probability distributions of the positions of markers fixed to the human body. The marker positions are measured by a 3-D motion capturing system. By utilizing this map, a robot can estimate human positions and postures from the tactile information. To verify the validity of our method, we conducted experiments using a robot covered with soft skin embedded with 46 tactile sensors. First, we collected data of human positions, postures and tactile information during an interaction between a human and the robot. Second, we built a map based on this information. Finally, we implemented our method allowing the robot to estimate the human's positions and postures.
In this study, we focus on a rehabilitation motion of human wrist joint and aim at supporting these rehabilitation trainings by introducing a mechanical system like a robot. A pneumatic parallel manipulator is introduced since it can drive enough D.O.F. to correspond to a wrist motion and has inherent compliance characteristics resulted from air compressibility. We propose a strategy where an equipent acquires the P.T.'s rehabilitation motion in order patient to receive rehabilitation motion through the equipment as if they are trained by P.T. Then we derive a method to estimate wrist impedance in multiple D.O.F. and display the wrist mechanical property in a real-time graphical manner. The effectiveness of the proposed rehabilitation strategy is veryfied through some experiments.
In general, robot should not be operated while it is in a singular configuration since operation degenerates its degrees of freedom. However, humanoid robots are capable of performing tasks while in a state of singular configuration because humanoid robots have many degrees of freedom. In this paper, new task method is proposed for humanoid robots, featuring selective utilization of their actuators while in the singular configuration. First, we define the effective singular configuration for a task and introduce a classification for joints. Next, we propose a method to identify the joint as effectively having a singular configuration. Finally, we demonstrate physical simulation results that include leading a robot in the singular configuration through the task. The advantages of the proposed method are that it supports economizing the robot's energy consumption and applies more power to the hand.