In this paper, we introdude a new control method of semi-passive dynamic walking for a biped walking robot, which uses actuators just only to maintain the passive walking. First, focusing on the contact phase of swing leg, the stability analysis of passive walking robot is performed through a discrete-time system of the robot. Second, we propose a new control method of semi-passive walking based on the delayed feedback control, which does not require any reference trajectory for walking. Finally, the effectiveness of the control law is shown through several sirriulations.
The purpose of this paper is to show the realization of passive walking of a biped robot Emu that is composed of 1 body and 2 legs. In the beginning of this paper, we fomulate the problem and analyze the stability of the walking with attitude control of a body. The walking is called “semi-passive walking”. We show, next, a numerical simulation on the semi-passive walking. Finally, we show some experimental results to verify a validity of the semi-passive walking and the analysis.
A new approach to the detection of the position and the orientation of planar motion objects based on one-sided Radon transform is presented. Detection of position and orientation of planar motion objects is a key to advanced object handling. First, one-sided Radon transform is introduced and its properties are investigated. Second, algorithms to detect planar motion of objects are constructed based on the properties of one-sided Radon transform. The algorithms are then implemented on a computer and are evaluated experimentally.
Multivalued logic is important as a fundamental technique in designing machine intelligence. Particularly Boolean multivalued logic such that the whole set of logic formulas forms a Boolean algebra inherits those theorems, laws, etc. which are obtained in traditional Boolean binary logic. In this article, only logic such that logic formulas take more-than-two truth values is called multivalued logic, and any logic such that weights or costs are added to logic formulas of Boolean binary logic is not classified into multivalued logic. This article defines Boolean multivalued logic by coding binary fractions being greater than or equal to 0 and less than 1 directly as truth values. To handle readily Bayesian theory rationalizing collection of knowledges via observation, this article also introduces an arithmetic operation called “conditioning” in addition to usual logic operations. A key to advancing machine intelligence built in a certain kind of robots required an ability of thinking is extracting causality between objects by introducing such a robust logic that can process inferences consistently. This article shows with some instances the way of optimizing truth values of atoms when what truth values some logic formulas should take are given as knowledges, and the way of calculating the truth values of unknown logic formulas as inferences. It also mentions possibility of introducing natural language for realization of phonic conversation between users and machine intelligence.
We developed an easy-to-operate, omni-directional vehicle including the power assist technology that acts for both the lon-gitudinal and rotational motions of the cart. We had two challenges during the development of this power-assisted cart. The first one was that, due to the relatively long body of the cart, it was difficult to keep the balance between lateral movement and turning movement of the cart when shifting the cart laterally. Therefore we modified the equation for calculating the cart turning speed so that the moment, which was driven by the operating force in the right/left direction, was offset. This way, we could stabilize the balance, resulting in a largely increased operating performance in the lateral direction. The second problem was that, during the one-hand pull-operation, the cart tended to run off its course to the right/left, especially during cornering. To solve this problem, we developed a positional control that detects the amount of lateral deviation by dead reckoning. Do-ing so, we could efficiently reduce the lateral deviation of the cart. We have succeeded in largely improving the operating performance by developing the above control technique.
A wave Central Pattern Generator (CPG) model is a mathematical model of nonlinear oscillators and the oscillator controls the movement of a leg. The wave patterns can be controlled by the virtual energy (Hamiltonian), and as the result the gait generation and the walk speed control are achieved. The toe trajectory is derived from the oscillator variables with based on some researches about the energy consumption in animals and robots' walking. The real robot experiment reveals the relation among Hamiltonian, the actual energy consumption and the walk speed, and then the effectiveness of the proposed method is verified.
Using the technique of impedance perception, which we previously proposed, the stiffness matrix which constrains the motion-force relation of the robot's end-effector is estimated on-line, and the uncertainties of the estimates are evaluated. This paper proposes a method of extracting information of surface properties, including normal direction, stiffness, and friction coefficient, from the estimated stiffness matrix obtained during dynamic friction situation on an environment surface. This technique can be implemented as an encapsulated perception function, and enables qualitative recognition of the environment structure. Results of preliminary experiments are presented.
In this paper we propose a novel method of sensor planning for a mobile robot localization problem. We represent causal relation between local sensing results, actions, and belief of the global localization using a Bayesian network. Initially, the structure of the Bayesian network is learned from the complete data of the environment using K2 algorithm combined with GA (genetic algorithm) . In the execution phase, when the robot is kidnapped to some place, it plans an optimal sensing action by taking into account the trade-off between the sensing cost and the global localization belief which is obtained by inference in the Bayesian network. We have validated the learning and planning algorithm by simulation experiments in an office environment.
During the behavior imitation, human being doesn't practice simple coordinates transformation, but acknowledge the others' behavior, understands the behavior after abstraction into symbol information, and generates one's self behavior. A framework “Mimesis” in cognitive science and “mirror neuron” found in biology field show that the behavior generation process isn't independent of behavior cognition process, but generation and cognition process have close relationship each other. Focusing on these facts, we propose a new method which carry out the behavior cognition process and behavior generation process at the sanie time, and co-evolve these two processes using the proto-symbol and mimesis framework. We also propose a mathematical model based on Hidden Markov Models in order to integrate the behavior cognition and generation process, which has an advantage that the model have three functions; 1) behavior memorization, 2) behavior recognition, and 3) self behavior generation, by itself. Finally, feasibility of this method is shown through experiment in a humanoid simulator.
In this paper, we extend the concept of dynamic manipulability measure to that for multi-fingered hands and propose a dynamic multi-fingered manipulability (DMM) measure. Unlike the conventional dynamic manipulability measure, the proposed measure considers internal forces and the dynamics of a grasped object. Specifying constant internal force components, we obtained a one-to-one relation from a joint torque to an object acceleration. The proposed measure is further extended to a measure for master-slave hands as a design index. Based on the new design indices, some two-fingered master-slave hand designs with different joint configurations are evaluated and an optimal link length is determined.
In this paper, a method for designing morphology of body and neural systems of link-type robots is suggested in which the robots can adapt the changes in environment using the evolutionary computation. The morphology of the body and neural systems have a close relationship to each other. So the model of the robot is constructed in which the morphology of the body and neural systems emerge simultaneously. The morphology of the body and neural systems are generated using a Genetic Programming. As a neural system, six kinds of them are used to compare with each other. The tasks are that the robots move on grounds including different height of hills from generation to generation in the two dimensional lateral simulated world under the effect of the gravity. The robots are evaluated based both on a moving distance and an efficiency. As a result, various combinations between the morphology of the body and neural systems of the robots were emerged. Moreover, the robot went over the hills that were not experienced. Finally, this method is applied to design the real.