This paper deals with position-based 6-DoF visual servoing. With a common sense of feedback control, we stress that improvement of the dynamics of the sensing unit is important for a stable visual servoing. We had proposed a method to improve dynamics in visual recognition, with compensating the fictional motion of the target in the camera images based on kinematics of the manipulator, by extracting the real motion of the target. We named it as hand-eye motion feedforword (MFF) method. In this paper, we present new visual servo system including MFF method; and confirm that the enhanced dynamics of recognition gave further stability and precision to the total visual servoing system, evaluated by full 6-DoF servoing experiment using 7-link manipulator. The convergence time of step response was about 10[s] and precise visual servoing to a moving target object has been achieved.
Caging is a geometrical method to constrain an object: a position-controlled robot surrounds the object to make it be inescapable from the cage composed by the robot bodies. In this paper, we propose three-dimensional caging by a multifingered hand (3D multifingered caging). In multifingered caging, force control is not necessary to constrain an object. This is an advantage over conventional grasping from the viewpoint of easy execution by actual robots. Furthermore, considering a margin of caging constraint allows the error of position control and deformation of objects to some extent. We classify 3D multifingered caging according to states of constraint, and derive sufficient conditions for caging of some simple-shaped objects. We show some configurations of a robot hand for caging planned with the sufficient conditions.
This paper proposes a novel pattern classification method of user's motions to use as input signals for human-machine interfaces from electromyograms (EMGs) based on a muscle synergy theory. This method can represent combined motions (e.g. wrist flexion during hand grasping), which are not trained by a recurrent neural network in advance, by combinations of synergy patterns of EMG signals preprocessed by the network. With this method, since the combined motions (i.e. unlearned motions) can be classified through learning of single motions (such as hand grasping and wrist flexion) only, the number of motions could be increased without increasing of the number of learning samples and the learning times for controlling of the machines such as a prosthetic hand. Effectiveness of the proposed method is shown by the motion classification experiments and prosthetic hand control experiments. The results showed that 18 motions, which are 12 combined and 6 single ones, can be classified sufficiently through learning of 6 single motions only (average rate: 89.2 ± 6.33%), and the amputee could control of a prosthetic hand using single and combined motions at will.
High mobility and large payload capacity are important abilities for wall climbing robots. Though conventional designs have achieved only either of these two abilities at the same time, a novel swarm type wall climbing robot system “Anchor Climber” that can achieve both of them is proposed in this paper. This robot system is composed of two or more child units and a parent unit, and wall climbing robots “Adhering Mobile Units” (AM Units) are used as these child units. AM Units can shift two adhering states for moving and for sticking on walls. Three concept models of AM Units using the technique of internally-balanced magnet (IB Magnet) were proposed, and two prototypes with omni-wheels and crawlers were developed. Several experiments were made with developed prototype robots, and suitable configuration and mechanisms of AM Unit were discussed from results of these experiments.
Traditionally, robots and humans work independently, isolated from each other for safety reasons. We developed a robot for use on the assembly line which works together with a human operator, assisting and augmenting the operator's skills. Because of its function, this robot is known as the “Car Window Installation Assist Robot” and is currently in use on the plant of Toyota Motor Corporation. In this paper, we propose a method to estimate human intent as well as a skill assist control method. Moreover, we present technology that enables safe human-robot cooperation, high product quality, and the cycle time of only 48 seconds.
This paper describes a method of robustly detecting and tracking road boundaries for autonomous navigation. Since sensory evidence for detecting road boundaries might change from place to place, multiple sensory features should be utilized. It is also necessary to cope with various road shapes. We develop a particle filter-based boundary tracking method. It makes use of various sensory evidence in the filter via respective likelihood functions and adopts flexible road models which are naturally generated in the particle filter framework. The proposed method has been successfully applied to various road scenes.
This paper presents a method to detect a road boundary using stereo images taken at on-vehicle stereo cameras. At first, an area whose height is different from a road plane is extracted by Planar Projection Stereopsis method. Next, boundary points between an obstacle and a road are detected and they are projected to a road plane. A road boundary is estimated in this projected images by the energy minimization method such as Snakes. The degree of a road boundary is calculated at each boundary point in a projected image. Road boundaries are estimated by searching a smooth line which passes on boundary poins whose the degree of a road boundary is high. Experimental results for real road scenes show the effectiveness of the proposed method.
This paper describes a local scan matching method that can estimate displacements of position and orientation between two successive scans fast, accurately and robustly. To estimate displacements of position and orientation, the conventional methods need to utilize nonlinear optimization method such as Levenberg-Marquardt method since there is no closed-form solution available which minimizes an evaluation function. Due to such nonlinear optimization method, the conventional method is so time consuming. In this paper, we formulate estimating displacements of position and orientation as a constrained least square problem and provide a closed-form solution which minimizes the evalution function defined by point-to-plane error metric. In the formulation, we show that a quartic function of Lagrange multiplier can be approximated by a linear function. These formulation enables our local scan matching method to be faster than the conventional method. The experimental results in three kinds of real office environment show the usefulness of our method.