This paper proposes a robotic assistance system for object handling based on imitative learning. At first, the system learns temporally short segments of motion called“motion primitives”from observation of human object handling tasks. Secondly daily human object-handling is recognized as a sequence of motion primitives. Then the occurrence of an appropriate assisting task defined as a sequence of motion primitives is predicted. Finally the corresponding assisting trajectory is generated from the sequence of motion primitives. The system is composed of such algorithms as object handling motion clustering, human motion recognition, assisting task prediction and trajectory generation, which are learned from human motion. On the other hand, the user specifies the tasks beforehand which the system should support. The validity of the proposed algorithms is confirmed through the experiment of object-handling assistance utilizing a cup.

View full abstract

Our goal is to realize a humanoid robot that has the capabilities of recognizing simultaneous speech. A humanoid robot under real-world environments usually hears a mixture of sounds, and thus three capabilities are essential for robot audition; sound source localization, separation, and recognition of separated sounds. In particular, an interface between sound source separation and speech recognition is important. In this paper, we designed an interface between sound source separation and speech recogniton by applying Missing Feature Theory (MFT) . In this method, spectral sub-bands distorted by sound source separation are detected from input speech as missing features. The detected missing features are masked on recognition not to affect the system badly. Therefore, this method is more flexible when noises change dynamically and drastically. It is the most important issue how distorted spectral sub-bands are detected. To solve the issue, we used speech feature apropriate for MFT-based ASR, and developed automatic missing feature mask generation. As a speech feature, we used a Mel-Scale Log Spectral (MSLS) feature instead of Mel-Frequency Cepstrum Coefficient (MFCC) which is commonly used for ASR. We presented a method of generating missing feature mask automatically by using information from sound source separation. To evaluate our method, we implemented it in a humanoid robotSIG2, and performed the experiments on recognition of three simultaneous isolated words. As a result, our method outperformed conventional ASR with MSLS feature.

View full abstract

We propose the real-time Q-MDP value method for decision making of a robot under uncertain state recognition. When the computation result of a control problem is known on the assumption that recognition is certain, the original Q-MDP value method decides an appropriate action based on uncertain recognition. The method is not suitable for real-time decision making due to the complexity of probability calculation. In the real-time Q-MDP value method, a particle filter that is utilized for state estimation is directly used for the probability calculation. The proposed method can make it possible to execute the Q-MDP value method in real-time. The proposed method is applied to total behavior of a goalkeeper for robot soccer competition. Experiments and actual games have suggested that this method can decide actions effectively according as uncertain result of state estimation.

View full abstract

A main purpose of our research is realizing an effective human-robot cooperation with physical interaction. In this paper, an architecture for the cooperation is proposed. As an example of the cooperation, ballroom dancing by a human and a robot is focused. And a dance partner robot referred to as “MS DanceR (Mobile Smart Dance Robot) ” is developed. MS DanceR realizes a ballroom dance together with a human based on a control architecture referred to as “CAST (Control Architecture based-on Step Transition) ”, which is designed according to features of ballroom dances. Experimental results illustate the validity of MS DanceR and its control architecture CAST.

View full abstract

In case of natural disaster like earthquake, Micro-Flying Robot (μFR) will be very effective for surveying the site and environment in dangerous area or narrow space, where human cannot access safely. In addition, it will be a help to prevent secondary disaster. This paper is concerned with modeling and autonomous control of μFR. However, since μFR is very small and light, sensors, such as GPS, cannot be carried. So the 3D position of μFR is measured by using one CCD camera to recognizing a marker attached to μFR. And we adopted the PID control which the performance can be raised by tuning even there is no model. But about the direction X and Y, it is not easy to find the gain of the PID controller, so at first we do the modeling by system identification and design the model base controller.

View full abstract

In this paper, we propose a robust online action recognition algorithm with a segmentation scheme that detects start and end points of action occurrences. Specifically, the alogorithm estimates reliably what kind of actions occurring at present time. The algorithm has following characteristics. (1) The algorithm incorporates human knowledge about relations between action names in order to toughen the recognition, thus it labels robustly multiple action names at the same time. (2) The algorithm uses time-series Action Probability that represents the likelihood of each action occurrence at every frame time. The Action Probability is obtained from time-series human motion using support vector machine. (3) The algorithm can detect robustly and immediately the segmental points using classification technique with hidden Markov models (HMIs) . The experimental results using real motion capture data show that our algorithm not only prevents the system from making unnecessary segments due to the error of time-series Action Probability but also decreases effectively the latency for detecting the segmental points.

View full abstract

We describe efforts to induce a quadruped robot to walk with medium walking speed on irregular terrain based on biological concepts. We so far reported our experimental results of dynamic walking on terrains of medium degrees of irregularity with a planar quadruped robot “Patrush” and a three-dimensional quadruped robot “Tekkenl”. What we discussed and experimentally examined in those studies was how to design sensorimotor coordination system for adaptive dynamic walking. In this paper, we make the definition of biologically inspired control and summarize how to construct the neural system while introducing the nervous system of animals, relating studies on computational neuroscience and robotics, and our former studies using Patrush and Tekkenl. We propose the necessary conditions for stable dynamic walking on irregular terrain in general, and design the mechanical system and the neural system by comparing biological concepts with those necessary conditions described in physical terms. PD-controller at joints constructs the virtual spring-damper system as the visco-elasticity model of a muscle. The neural system model consists of a CPG (central pattern generator), reflexes and responses. We add several new reflexes and responses in order to satisfy the necessary conditions for stable dynamic walking in outdoor environment. We validate the effectiveness of the proposed neural system model control by making a self-contained quadruped robots called “Tekkenl” walk on natural ground. Consequently, we successfully propose the method to integrate CPGs and sensory feedback for adaptive dynamic walking of a quadruped.

View full abstract

In this paper, we propose a task skill transfer method using a bilateral teleoperation system. A target area of the task skill is a practical and a dexterous task that is performed by a manipulator. The task skill is implemented by an impedance control. In our method, an operator executes a task under a bilateral teleoperation. He/she finds a suitable motion strategy for a robot. Then, a task skill is written by the suitable motion strategy of the task and the teleoperation results. A knock of a task realization is generally described qualitatively information by a person. However, the knock can be described quantitatively information (teleoperation result) by our method. Furthermore, we discovered that unconsciousness movement of a person became an important element of a knock of a task realization. To demonstrate our task skill transfer method, we choose a lever handle valve manipulation task. This demonstration shows that a task skill of a lever handle valve manipulation is able to be programmed easily, needs no kinematics models of the task, has a robustness of positional errors and has a performance ability and time as the same as the operator.

View full abstract

This paper discusses a design approach on robotic hand by considering the characteristics of visco-elasticity. “NORIMAKI” is a typical example where it has the visco-elastic characteristics. We first show that the dynamic characteristics of “NORIMAKI” can be expressed by utilizing the Maxwell model with two layers. Based on dynamic parameters obtained by experiments, we make clear the relationship among the total working time, the plastic deformation of food after the operation, the hand stiffness, and the operating velocity of the hand. We newly found an interesting behavior where there exists an optimum set of the design parameters for achieving the minimum plastic deformation.

View full abstract