We have developed a humanoid robot HRP-1S that can be controlled its whole body motion. In phase one of Humanoid Robotics Project (HRP) of METI, Honda R&D Co. Ltd. has produced humanoid robots HRP-1 as humanoid research platforms to establish good applications of humanoid robots in phase two. HRP-1 is controlled its feet and arms separately, so that it is not suitable for some applications. By modifying the hardware of HRP-1 and implementing our own control software on it, we made HRP-1S. In this paper, we present a control system of HRP-1S. Some experimental results are also shown.
This paper discusses the analysis of the ZMP (Zero Moment Point) of a humanoid robot under coordination of two arms and two legs. We show that there are two kinds of ZMPs for such coordination tasks, i.e., the conventional ZMP considering all sources of the force/moment acting in the foot supporting area, and the“Generalized Zero Moment Point (GZMP) ”which is a generalization of the conventional ZMP. By projecting the edges of the convex hull of the supporting points onto the floor, we show that the position and the region of the GZMP for keeping the dynamical balance can be uniquely obtained. We also show that there are edges of the convex hull where the robot can roll around the edge. The effectiveness of the proposed method is shown by some simulation results.
This paper presents a humanoid robot that has the same size as a human and that can lie down to the floor and get up from the floor with the robot face upward and downward. We believe that the robot is the first life-size humanoid robot with the capability. The motions are realized by the combination of novel hardware and software. The features of the hardware are a human-like proportion and joints with wide movable ranges including two waist joints. The software segments the motion into the sequence of the contact states between the robot and the floor and assigns an appropriate controller to each transition between the consecutive states. The experimental results are presented.
A human operator can operate a robot intentionally using remote control devices. Therefore, the remote control system has the advantages that the robot can work according to the circumstances. We applied it to tele-operation of a humanoid robot that drives industrial vehicles. We have developed remote control devices that can be transported easily near the robot and remote control methods to operate the whole body of the robot, which consist of supervisory control method and master-slave control method and the operator select them suitable for the situation. For the evaluation, the tele-operated humanoid robot: HRP-1S could succeed in driving a rift truck in a standing posture and driving a backhoe in a sitting posture.
A snake robot is a typical example of robots with redundant degree of freedom. Using input-output linearization for only movement of the head of a robot, we can control the head speed as a desired one, but eventually the robot will come to a singular posture like a straight line. In order to overcome the problem, a control with dynamic manipulability was proposed. In this paper, we propose a control technique in which a physical index of horizontal constraint force is used, and a control law for head configuration. By using these, the winding pattern with which the robot can avoid the singular posture is generated automatically and head converges to the target.
Imitation is an important capability for an intelligent robot to perform a variety of complicated tasks in the real world. It is one of the most interesting issues as a study of not simply accelerating behavior learning but also modeling the mechanism underlying human intelligence. From these viewpoints, a method for imitation based on the demonstrator's view recovery is proposed, in which the learner recovers the demonstrator's view that can be regarded as observing itself in its own view by assuming that the demonstrator's body structure is the same as the learner. The method is based on the opt-geometric constraint called “epipolar geometry” between the both views and on the adaptive visual servoing to follow the desired trajectory corresponding to the recovered demonstration in the learner's view. In order to estimate the parameters of epipolar geometry, we assume that the both initial postures are the same. It is shown that the learner can perform view-based imitation by the proposed method in the real robot experiment. Finally, we discuss our future work on imitation in the paradigm of demonstrator's view recovery.
In this paper, we describe a new way of attitude determination and motion planning of robotic architecture avoiding structural failure by restraining total strain energy stored in the architecture. In the proposed motion-planning algorithm, when the risk of member yielding becomes higher while the robot is in motion, a safer attitude for restraining its total strain energy is searched. At the next step, a new trajectory is created, beginning with the obtained attitude, and ends with the final target which is initially given. The procedure is repeated until a converged attitude is obtained. Risk for member yielding and total strain energy are calculated by using the Finite Element Method (FEM), and an attitude for restraining total strain energy is searched by using the Direct Search Method (DSM) . Some numerical tests are carried out with a truss-type robotic architecture and a 3-link manipulator, and interesting results are obtained by changing the judgment and target levels of structural parameters.
Although an environment map provides essential information for mobile robot navigation, an accurate map needs a lot of building cost and is not flexible to changes in object poses in the environment. A solution of these problems would be a framework of navigation using an inaccurate map. We have already proposed a representation of an inaccurate map and a localization method on the proposed map. In this paper, we present an object-recognition method suitable for mobile robot localization, and a path-tracking method on the proposed map. The robot recognizes objects such as desk and door, and localizes itself based on the relative poses from the objects. Since a path may also be inaccurate on an inaccurate map, the robot corrects the path based on localization results while tracking the path. Integrating these methods, we have built an indoor navigation system. An experiment shows that the robot successfully navigated in an indoor environment, recognizing several kinds of objects.
This paper is concerned with a method of identification of terrain shape and torque compensation using inverse model and a learning method for moving operations of mobile manipulators traveling on unknown irregular terrain. When the mobile manipulator operates during traveling on irregular terrain whose shape is unknown, the hand motion of the mobile manipulator is influenced by the terrain dynamically and kinematically. Therefore, both of dynamical and kinematical compensations are required so as to perform accurate operations for an object in the external world. In this paper, a new compensation method using a learning method is proposed. First, the validity of the proposed model is conformed to simulate the real model motion by comparing experiment and simulation results. Second, it is shown that influences of dynamical disturbance torques occur when the mobile manipulator travels fast on irregular terrain. Third, a torque compensator using GA (Genetic Algorithm) is proposed by referring the inverse model of the mobile manipulator. Finally, it is clarified that the proposed compensator can improve trajectory-tracking control performance by identifying unknown irregular terrain shape and performing torque compensations.
In the environment that human and robots coexist, mobile robots have to approach human as much as possible in order to provide human with physical services. For friendly interaction with human, the robots should be able to afford human-affinitive movement. In this paper, human following, as one of human-affinitive movements, of robot is achieved. The human following robot requires several techniques; the recognition of the target human, the recognition of the environment around the robot, and the control strategy for stable human following. In this research, Intelligent Space, where many sensors and intelligent devices are distributed, is used for the recognition of the target human and the environment around the robot. The control law based on the virtual spring model is proposed to mitigate the difference of movement between the human and the mobile robot. The proposed control law is applied to Intelligent Space and its performance is verified by computer simulation and experiment.
The purpose of this study is to develop a simple but useful robot with passive joints. As an example of such a robot, this paper presents dynamic rolling of a 5R closed kinematic chain, only two of whose joints are actuated.When it rolls, it has one DOF for its absolute orientation besides two DOFs for its shape. The absolute orientation is independent of the constraint for forming the closed kinematic chain. If its unactuated joints were actuated, the closed form would be over-actuated but the absolute orientation could not be driven directly. We show that the absolute orientation is subject to an acceleration constraint, not a velocity constraint. Therefore, the formulation of the dynamics of the rolling motion makes it possible to control the absolute orientation. The shape and orientation of the robot cannot be controlled simultaneously. This paper proposes a control strategy switching the shape and orientation controllers. The orientation controller reduces a negative acceleration caused by gravity which acts on the robot. We verify the control strategy by computer simulations and experiments.
In reinforcement learning, it takes long time to learn purposive behaviors due to the nature of delayed reward. Multiple reward functions are often introduced in order to accelerate the learning speed of obtaining complicated behaviors. However, the methods of the weighted sum of reward functions often cause the undesirable side effects because the objective functions are different from the original one. In this paper, we propose a novel hierarchical reinforcement learning method for utilizing multiple reward functions. The value function of the upper layer is estimated using the reward for accomplishing the entire task and the supplementary reward calculated from the value functions of the lower layers. The proposed method was applied to a simplified arm movement problems, and outperformed conventional methods.
This paper studies basic mechanisms of passive dynamic walking from the mechanical energy point of view, and proposes novel gait generation and control methods based on the passive dynamic walking. Firstly, we show the uniform property of passive dynamic walking that, the walking system's mechanical energy increases proportionally with respect to the position of the system's center of gravity, which results in an interesting indeterminate equation that determines the relation between the system's control torques and its center of gravity. We then show that active dynamic walking on a level can be realized by solving this equation for the control torques. Furthermore, the applications to robust energy referenced control, as well as the extension for controlling a kneed biped system are discussed. The effectiveness of the proposed methods have been investigated through numerical simulations.
This paper proposed an adjustable parallel mechanism, which can change its workspace by change of link parameters. Firstly, adjustable parameters and adjusting methods are classified and discussed. Secondly, the workspace of HEXA type with adjustable middle links is numerically analyzed. The mechanism is changed translationally and rotationally and the volumes and centroids of its translational and rotational workspace are compared. The analyses conclude that adjusted mechanism is superior to non-adjusted mechanism in terms of workspace. Finally, a prototype of an adjustable 5-bar-link mechanism is introduced for an instance of simple adjustable parallel mechanism.