International Journal of Humanoid Robotics

An apprentice is an able-bodied individual that will interactively assist an expert, and through this interaction, acquire knowledge and skill in the given task domain. A humanoid apprentice should have a useful repertoire of sensory-motor acts that the human can command with spoken language, along with a real-time behavioral sequence acquisition ability. The learned sequences should function as executable procedures that can operate in a flexible manner that are not rigidly sensitive to initial conditions. Our study integrates these capabilities in a real-time system on the HRP-2 humanoid, for learning a cooperative assembly task. We previously defined a system for Spoken Language Programming (SLP) that allowed the user to guide the robot through an arbitrary, task relevant, motor sequence via spoken commands, and to store this sequence as re-usable macro. Here, we significantly extend the SPL system: It integrates vision and motion planning into the SLP framework, providing a new level of flexibility in the actions that can be created, and it allows the user to create "generic" functions with arguments (e.g. Give me X), and it allows multiple functions to be created.

This paper presents a cooperative control software architecture that coordinates a team of multiple humanoid to complete a mission by collaborating with each other. The mission of the humanoid team is decomposed into tasks and distributed to each humanoid to be executed. Each task is described by the proposed humanoid action primitives, which are designed to abstract broad classes of humanoid tasks appropriately. In particular, missions and tasks for the humanoid team are designed by using a finite state machine with a developed user interface. The multiple humanoid cooperative control software consists of 3 layers: the mission layer, task layer, and action layer. The software architecture has scalability to the number of humanoids and the number of assigned missions with its framework based on the CORBA middleware, which integrates many different functionalities of the humanoid. The feasibility and robustness of the implemented software architecture are verified through successful completion of the mission assigned to the humanoid team while each humanoid performs its given task sequentially.

In this paper, dynamic stair climbing and descending are experimentally realized for a biped humanoid robot, HUBO. Currently, in addition to biped walking on the ground, other types of biped walking such as running, jogging, and stair walking (climbing and descending) have been also studied since the end of 1990. In spite of many years of research works on stair walking, it is still a challengeable topic that requires high performance of control technique. For dynamic stair walking, we designed stair climbing and descending patterns according to a known stair configuration. Next, we defined stair climbing and descending stages for a switching control strategy. In each stage, we designed and adopted several online controllers to maintain the balance. For the simplicity and easy application, the online controllers only use the force and torque signals of the force/torque sensors of the feet. Finally, the effectiveness and performance of the proposed strategy are verified through stair climbing and descending experiments of HUBO.

This paper discusses the generation of a running pattern for a biped and verifies the validity of the proposed method of running pattern generation via experiments. When a running pattern is created with resolved momentum control, the angular momentum of the robot at the Center of Mass (COM) is set to zero, as the angular momentum causes the robot to rotate. However, this also induces unnatural motion of the upper body of the robot. To resolve this problem, the biped was set to a virtual under-actuated robot with a free joint at its support ankle, and a fixed point for a virtual system was determined. Following this, a new periodic running pattern was formulated using the fixed point. The fixed point is easily determined using a numerical approach. In an experiment, the planar biped ran forward using the proposed pattern generation method for running. Its maximum velocity was 2.88 km/h. In the future, faster running of the biped will be realized in a planar plane and the biped will run in an actual environment.

Transferring motion from a human demonstrator to a humanoid robot is an important step toward developing robots that are easily programmable and that can replicate or learn from observed human motion. The so called motion retargeting problem has been well studied and several off-line solutions exist based on optimization approaches that rely on pre-recorded human motion data collected from a marker-based motion capture system. From the perspective of human robot interaction, there is a growing interest in online motion transfer, particularly without using markers. Such requirements have placed stringent demands on retargeting algorithms and limited the potential use of off-line and pre-recorded methods. To address these limitations, we present an online task space control theoretic retargeting formulation to generate robot joint motions that adhere to the robot's joint limit constraints, joint velocity constraints and self-collision constraints. The inputs to the proposed method include low dimensional normalized human motion descriptors, detected and tracked using a vision based key-point detection and tracking algorithm. The proposed vision algorithm does not rely on markers placed on anatomical landmarks, nor does it require special instrumentation or calibration. The current implementation requires a depth image sequence, which is collected from a single time of flight imaging device. The feasibility of the proposed approach is shown by means of online experimental results on the Honda humanoid robot — ASIMO.

This paper presents an applicable formula for determining the workspace of digital human lower extremities. The digital human model has over 100 degrees of freedom (DOF): 94 in the upper body, 14 in the lower extremities, 5 in the neck, 4 in the eyes, and 25 for each hand. The Jacobian row rank deficiency criteria are implemented to determine the singular surfaces that finally form the workspace. The use of this digital human model for determining workspace offers several advantages over direct measurement: (1) the workspace can be visualized in real-time based on offline computation, (2) the workspace can be used for the ergonomic design of products in the virtual prototyping stage, and (3) the calculated workspace includes complete information about the envelope and inside characteristics.

There are many interests in developing life-like robots, or robots which are both intelligent and autonomous. And, one obvious characteristics of life-like creatures is that they can autonomously develop and learn during their life span. Such abilities obviously depend on the ways of designing human-like minds. Then, a fundamental question is how to devise the innate, or built-in, principles behind the blueprint of a human-like mind, and to apply these findings to guide the design of the mind of life-like robots. In the literature, there are two schools of thoughts. One advocates the study of the nervous systems of biological brains (e.g. human brain) until the discovery of the blueprint of a mind. The second approach is to follow the path of invention and validation until the full understanding of physical principles which enable the design of an artificial mind that is as good as a biological mind. This paper embraces the second approach, and aims at formulating a new ground which could guide the design of the minds of life-like robots at various stages. In particular, the discussion is focused on answering the question of what life is from an engineering point of view. And, we approach the answer by examining the key steps of evolution from non-life to life. In this paper, five key steps of evolution from non-life to life will be discussed in detail. They are embodiment of energy flow, embodiment of signal flow, embodiment of knowledge flow, embodiment of decision flow, and embodiment of awareness flow. These findings are grounded on our engineering works toward the development of low-cost humanoid (LOCH) robot, and offer a unique perspective and an engineering basis. Whenever possible, the discussions in this paper are supported by real results of experiments on real robots.