In a grinding robot system, a grinder, which is mounted on the wrist of the manipulator, is liable to vibrate and sometimes made unstable because the stiffness of the manipulator is insufficient. The vibration makes the ground surface rough and the unstable condition may break the manipulator.
The stability in grinding using a compliant manipulator is discussed in this paper. If a plane surface of a workpiece is ground, the motion of the grinder is represented by a linear differential equation. From the equation, it is shown that the stability of the grinder depends on the stiffness matrix of the manipulator, an angle of the inclined plane and a coefficient of friction between the grinder and the workpiece. When a curved surface is ground, the differential equation turns to nonlinearity. Then the bounds of the stable condition is affected by the curvature of the ground surface. From the discussion, it is shown that the rotational angle of the manipulator wrist must be controlled to keep the grinder stable in grinding of a curved surface.

View full abstract

A procedure for synthesis of multi-degrees of freedom mechanisms composed of connecting chains has been developed. The concept of mobile and quasi-mobile screw spaces comprised by joints in the connecting chains was newly introduced in this paper. When an output link and a fixed link are connected by plural of connecting chains, the output link can make any motion expressed by an arbitrary screw in the intersection of the screw spaces of the connecting chains. It is proved that the sufficient condition that the intersection of screw spaces of the connecting chains does not change in the entire range of motion of the mechanism is for each screw space of the connecting chains to be mobile or quasi-mobile with respect to the motion of the output link. Examples of synthesis of mechanisms for translational motions show that a multi-degrees of freedom mechanism which can make specified motions can be synthesized by using this condition.

View full abstract

In hand-eye systems for advanced robotic applications such as assembly, degrees-of-freedom of the vision sensor should be increased appropriately to cope with unstable scene conditions. Particularly, in case of using a simple vision sensor, an intelligent use of the sensor is essential to compensate for its inability to adapt to changing environment. This paper describes a vision sensor set-up planning system which automatically generates plans, based on the environmental models, of illumination positions to achieve reliable tasks. A typical vision task in which edges on an object are measured to determine the position and the orientation is assumed for the sensor set-up planning. In this context, the system is able to generate plans for camera position, illumination position, and a set of edges to be measured. The system determines the best plans of illumination for stationary or moving objects by evaluating scene conditions for reliable measurement of the object position such as edge length, contrast, and relative angles based on the model of the object and the task environment. Automatic vision sensor set-up planning functions as shown in this paper will play an important role not only for autonomous robotic systems but also for teleoperation systems in assisting advanced tasks.

View full abstract

In biped locomotion, there are double legs supporting phase with kick and single leg supporting phase with making a forward step. The purpose of this study is dynamic analysis and control in double legs supporting phase.
This phase is a closed loop links system (constraint dynamic system), in which dynamic states (floor reaction, joint torque) are restricted because the foot are not fixed on the floor. Therefore, it is preferable to use an analytical method by which dynamic states would become clear.
In this respect, this paper first refers to analytical method by D'Alembert's principle, in which static forces are derived from static equilibrium equations for each link. Then, based on analytical result, control experiment is made, and the utility of this method is shown.

View full abstract

The closest points on a robot and its obstacles are quite important for generating robot motions to avoid the obstacles. Almost all algorithms previously presented for determination of the closet points cannot work well in three-dimensional cluttered space which has many obstacles with complicated shape since their computational times depend on shape complexity of the robot workspace, that is, the total number of vertices of polyhedra representing the obstacles. In view of this, a feasible algorithm that runs on the octree representation integrating all obstacles is proposed in this paper. Due to a hierarchical structure of the octree, the algorithm can investigate a part of the obstacles in order of the distance from the robot and consequently its computational time does not directly depend on shape complexity of the workspace. In result, the proposed algorithm can fulfill its function even in any cluttered space. Finally, experimental results are presented to show that the algorithm runs fast in the cluttered space.

View full abstract

Since the friction between a finger tip and object surface is not taken into consideration in conventional stability analyses of multi-fingered hand based on stiffness model, a significant contradiction occurs when making these results apply to a practical grasp system. For example, even though the stability is guaranteed for a grasp system without friction, there is a specific case that the grasp system becomes unstable for a small rotation of an object under frictioned model.
This paper discusses a stability of grasp system with friction by substituting two perpendicular virtual springs k_xi and k_yi going through the same contact point for each finger. The rotational center is predetermined within an object and the direction of the virtual spring k_yi is so selected that the extended line passes through the center. Stiffness matrix of grasp system is first introduced by synthesizing all virtual springs, and the stability condition is made clear by exploring the positive definite of the matrix. It is shown from this analysis that virtual spring k_yi plays an important role for generating internal force of an object in an equilibrium state while k_yi is indispensable to the stability of rotation. An evaluation function, which is specified the maximum internal force in an equilibrium state, is also introduced to keep off an excessive force to the object when in actuality determining the grasp parameters such as virtual stiffness, initial compression value of virtual spring. Finally, the validity of the analysis is confirmed through some basic experiments using two-fingered hand model which is specially designed for this study.

View full abstract

This reseach is concerned with impedance control of a manipulator which exerts stable contact task.
The method controls dynamic interaction between a robot and its environment by changing the apparent mechanical impedance of the manipulator.
Previous impedance control methods required force or torque sensors, which made the manipulator system most complex.
In this paper a new method is proposed to control the impedance of a manipulator without using force or torque sensors. Angular velocity and acceleration of the manipulator joints are estimated, and by using the computer model of the manipulator necessary torque for each joint is caluculated and applied to the joint to attain the desired impedance. The feasibility of the method is verified by the surface following experiments and collision experiments using a two degree of freedom DD manipulator.

View full abstract

A new learning control method with a current signal of a direct current motor (D. C. M.) is proposed. There are two merits in the new learning control method. Firstly, the current signal of the D. C. M. is easily obtained compared with the acceleration signal. Secondly, since the current signal includes nonlinear terms such as centrifugal force, Coriolis force and so on, very high convergence rate of learning is obtained. The current signal is so similar to the torque signal that we need nominal values of parameters to learn the current signal. But this method is still a learning control basically which obtains the optimal input by repetitive operations and modifications. Therefore, we do not have to estimate the real values of the parameters. In short, this learning method attains high convergence rate by using the nominal values of the parameters which we can obtain easily.
In this system, an armature resistance Ra of a D. C. M. and the physical parameters of the robot arms are difficult to identify exactly. Then, firstly we examine relations between the variation of Ra and the learning stability. Secondly, we show that the learning gain of the current signal should be reduced after the trajectory converges to some extent, because the estimation errors of the physical parameters of the robot arms degrade the learning convergence. Finally, the proposed algorithm is shown to have high convergence rate compared with the one with the position, velocity and acceleration signals. The proposed algorithm is advantageous especially when the nonlinearity of the robot arms is strong.

View full abstract

We have already proposed an identification method of inertial terms of a robot arm based on the mass loading effect of mechanical resonance. This method assumes that the mechanical cramp is prepared for every joint axis except for the measured joint. Although the measurements are performed without mechanical cramp.
Analyzing the effect of identification accuracy under the condition of non-mechanical cramp, it is clarified that some estimation errors due to the coupling vibration occur.
In this paper, we propose a modified identification method of inertial terms with the positive feedback of velocity and acceleration.

View full abstract

This paper describes the representation of map and generation of perspective image for mobile robot. This map gives an environment model for already proposed mobile robot vision system for reckoning in the building environment using color image.
Requirements for the map are:
1) Map must include parameters of vertical direction to generate perspective image based on estimated robot position.
2) Map must represent connected corridors and rooms.
3) Map must represent the attributes of the door opened. Representation of map is given by S-expression like text. It consists of three lists
(a) a list for definition of primitives representing objects in the building environment.
(b) a list for definition of arrangement of primitives in a segment and for definition of segment location. A segment represents a corridor or a room.
(c) a list for indicating open angle of doors.
In a primitive definition list, a pillar primitive, a dent primitive, a door primitive, a cylinder primitive are provided for representation of objects (these are called form primitive) . A rectangular color primitive and a cylinder color primitive are provided for representation of colored part on objects and they are attached on form primitive.
Generation of perspective image from map is performed by seeking for characteristic points on the color primitive and projection of these points on the image plane with giving visual angle of the camera and robot position and direction on the map. Then colored objects are represented by trapezoids. As objects in the real image are also fitted to trapezoids in the proposed vision system, matching investigation is performed by comparison of trapezoids from map and from real image.
An experimental example of generated perspective image is also shown.

View full abstract