International Journal of Humanoid Robotics

This paper deals with the mechanical design, system integration, and dynamic walking algorithm of KIBO, an emotional biped humanoid robot that has a facial expression mechanism and various human-interactive devices. To emphasize the aesthetic features and marketability of KIBO, the mechanical design was performed after the exterior design stage to conform to all requirements, particularly constraints imposed by the external appearance and human-like link dimensions. For flexible biped walking, a walking pattern generator with variable walking parameters was developed. The walking pattern generator generates both a walking pattern and a corresponding reference zero-moment point (ZMP) pattern simultaneously. For stable biped walking, a walking control strategy using the ZMP and inertial sensor data was developed. In the strategy, we newly proposed a dual ZMP control approach and a posture control approach using an equivalent body inclination, which is calculated from the ZMP and inertial sensor data for robust walking on non-level ground. Finally, the hardware, software architecture, and dynamic walking performance of KIBO were verified through several walking experiments.

In this paper, we propose and evaluate methods for the local navigation using only visual perception for the skiing robot. Our skiing robot, capable of skiing using the carving technique, has no direct control on the velocity of skiing as it cannot break or accelerate, therefore well known navigation methods for nonholonomic mobile robots cannot be directly applied. We consider the following methods: an intuitive method of aiming at the closest gates, a human obstacle avoidance movement model, neural networks learning from a set of human demonstrations, and a global method that uses a predefined, spline-encoded path. The navigation performance of the robot on unknown ski courses is evaluated using two criteria: successful completion of the course and the time required to complete the course. Simulation results show the applicability and drawbacks of presented methods. Finally, the method using the neural networks was applied on a real-world skiing robot and we tested navigating a slalom course on both roller blades and skies.

Application of Jacobian determinant of a nonredundant serial chain robot is a powerful and elegant way to deduce a closed-form computation of its working volume and the structural robot optimization. This approach highlighted in the 80's seems to have been neglected since then. We discuss under which joint conditions it can be considered and we show the possibility to extend this approach to redundant serial chain robots by substituting nonsquared Jacobian minors for squared-Jacobian determinant, thanks to the powerful so-called cubical dissection theorem peculiar to the zonotope theory. Two examples are considered: a 4R anthropomorphic regional arm structure and a 3R planar human digit-like structure.

More and more systems are developed that include several robot arms, like humanoid robots or industrial robot systems. These systems are designed for complex tasks to be solved in cooperation by the robot arms. However, the capabilities of the individual robot arms to perform given tasks or the suitability of a multi-robot system for cooperative tasks cannot be intuitively comprehended. For planning complex tasks or designing robot systems, a representation of a robot arm's workspace is needed that allows to determine from which directions objects in the workspace can be reached. In this paper, the capability map is presented. It is a representation of a robot arm's kinematic capabilities in its workspace. The capability map is used to compare existing robot arms, to support the design phase of an anthropomorphic robot arm and to enable robot workcell planning.

This paper explores the vertical upward jumping of a planar biped. There are two stance phases and one flight phase in the jump. One stance phase takes place before the flight phase, another one after the flight phase. The stance phase before the flight phase is decomposed into several parts: A crouching, a thrust at the knees, a rotation of both feet (massless) around their toes. The second stance phase (after the flight phase) starts with a touchdown of the toes. It consists of a feet rotation, a touchdown of the whole soles and finally of a straightening up movement of the biped. A mathematical model for this kind of jump is developed. Torques are applied at the hip, knee and ankle joints. The control algorithm is designed to ensure the jump of the biped. The synthesis of the jumping process is supported by simulation, which gives consistent results with human data from biomechanical literature. The stick diagram of the jump derived from simulation results seems natural for the human jumping.

This paper investigates the use of functional principal component analysis (FPCA) for automatic recognition of dynamic human arm gestures and robot imitation. FPCA is a statistical technique of functional data analysis that generalizes standard multivariate principal component analysis. Functional data analysis signals (e.g., gestures) are functions that are considered as observations of a random variable on a functional space. In particular, FPCA reduces the dimensionality of the input data by projecting them onto a finite-dimensional space spanned by a few prominent eigenfunctions. The main contribution of this work is the proposal of a novel technique for unsupervised clustering of training data and dynamic gesture recognition based on FPCA. FPCA has not been considered in previous studies on humanoid learning. The proposed approach has been evaluated in two experimental settings for motion capture. In the first setup single arm gestures are recognized from inertial sensors attached to the arm of the user. In the second setup the method is extended to two-arm gestures acquired from a range sensor. Recognized gestures are reproduced by a small humanoid robot. The FPCA method has also been compared to a high performance algorithm for gesture classification based on dynamic time warping (DTW). The FPCA algorithm achieves comparable results in both recognition rate and robustness to missing data, while it outperforms DTW in terms of efficiency in execution time.