Kyuhei Honda

Real-time pose estimation of an object manipulated by multi-fingered hand using 3D stereo vision and tactile sensing

We describe a real-time model-based sensing system to estimate the pose of an object manipulated by a multifingered hand in a realistic environment. The system uses the least-squares method to estimate the pose, on the assumption that the approximate initial pose and geometric shapes of the object and fingertips are known. 3D measurement of features on the surface of the object is achieved by visual tracking with stereo camera using a template matching method. The primary features used for visual tracking are distinguishable texture patterns on the object surface. In complex environments, visual tracking can be failed due to the various occlusion. The system solves the occlusion problem by detecting the occlusion and using contact information of fingertips with the 3D visual information. Actual implementation of the system is explained together with results of experiments about the stability of tracking, the accuracy of estimation, and the processing time of the system.

Detection and measurement of fingertip slip in multi-fingered precision manipulation with rolling contact

This paper describes a new method of detection and measurement of fingertip slip on the surface of a manipulated object in a multi-fingered precision manipulation with rolling contact. Due to the fact that slip sensors are still in a basic research phase and no slip sensor is compatible with the actual fingertip manipulation, our method is based on the multi-sensor fusion of vision, joint-encoders of the fingers, and fingertip force/torque sensors. Position of the contact between an object and the fingertip is computed in realtime from the measured pose of the object and the fingertip position. Tangential slip displacement of the fingertip contact is then separated from the displacement by rolling motion of the fingertip. Reliability of the slip detection from the noisy data is improved using the tangential-to-normal ratio of the measured force at the contact and the mutual relationship between the tangential force and the direction of the contact displacement.

Mechanical assembly based on motion primitives of multi-fingered hand

Summary form only given. The system we propose is a bridge between a hardware-oriented control system and a task-level description of mechanical assembly. Writing an executable program for the multi-fingered hand requires wide varieties of expertise. To cope with these problems, we propose a task execution system based on motion primitives for our multi-fingered robot hand system. The robotic mechanical assembly process is defined as a sequence of achieving different intermediate states of contact among mechanical parts until parts are fixed as specified. When a task is given, a programmer divides the motion of the manipulated object into several motion primitives each of which has a particular target state to be achieved in the task context. He can then describes a task program as a sequence of these motion primitives without knowing a specific hand configuration. We have developed a manipulation system with multi-fingered hand and multi-sensors. The vision system is composed of a pair of CCD cameras, a video input board and a video rate convolution board.

A dexterous manipulation system with error detection and recovery by a multi-fingered robotic hand

Describes a dexterous and robust manipulation system with a multi-fingered hand. The system is composed of a real-time pose estimation of an object manipulated by the hand, a precise manipulation using fingertip with rolling contact, motion primitives for mechanical assembly, and error-recovery strategies based on the task observation function. The system architecture and experimental results are shown.

Dexterous manipulation system with a multi-fingered hand and multi-sensors

This paper describes a dexterous manipulation system with a multi-fingered hand and multi-sensors. The system is capable of dexterous manipulation by rolling contact between fingertips and a grasped object. A mathematical model of the kinematics of rolling contact, a trajectory and force planning algorithm, a 3D real time visual system, and the actual implementation of the system are explained together with results of experiments verifying our approach.

Walking mechanism for a lion-type robot

This paper describes a new type of locomotion for a quadruped robot. Recently, many biomimetic quadruped robots have been developed. Most of them use many actuators for running, which causes them to become big in size. We have developed a medium-sized quadruped robot by using only one electric motor as an actuator for walking and running. We also adapt to the robot the mechanism which is similar to Theo Jansen mechanism to make its gait biomimetic. In order to realize our mechanism, we propose “Modified Gallop Gait” that is derived from bounce gait and gallop gait. The experimental results show that the robot runs as the modified gallop gait at the velocity of 0.89 m/s, while its pitch angle oscillates periodically to realize biomimetic motion and its stable running.

Mechanical assembly by a multi-fingered robotic hand with error recovery function

Describes a dexterous and robust manipulation system for mechanical assembly with a multifingered hand. The system is composed of a grasp planner and a real-time execution subsystem. The planner plans an initial grasp from which a required continuous motion of the object is achieved while keeping rolling contact at the fingertips. Task error detection and recovery strategies have been developed based on the task context. The system architecture and experiment results are shown.

Recognition of Hand Contact State with an Omnidirectional Camera and Laser Modules

In this paper, we propose a human-machine interface to recognize the users finger motion in order to perform object manipulation with a multi-fingered robot hand. The information needed for object manipulation includes the desired contact positions between robot fingers and the manipulated object as well as whether or not each finger should touch the object. Our system obtains the above information by using an omni directional camera and laser modules, while the user holds the camera with fingers by regarding the camera as a virtual manipulated object. After the features are extracted from the image which contains laser light reflecting on the users fingers, the features are used for a support vector machine to recognize the users contact state. The experimental results show that our system recognizes the contact state in real-time at more than 92 percent accuracy rate.