Koh Hosoda Pengarang

Miniature direct-drive two-link using a micro-flat ultrasonic motor K. Miyoshi and T. Mashimo Depar tment of M echanical Engineering, Toyohashi Universit y of Technology, Toyohashi, Aichi, Japan ABSTR AC T We propose a micro-flat ultrasonic motor that performs a high torque density by a tiny and flat stator for driving a miniature two-link. Such a two-link can be a fundamental technology that enables micro-assembling and micro-manipulation under a microscopic environment. In this paper, we build the micro-flat ultrasonic motor comprising of a flat stator with 2.6 mm in width and height and 1 mm in depth and evaluate the output performance measures. The prototype micromotor shows a torque of 55 µNm at a voltage of 80 Vp−p by optimizing the preload between the stator and rotor. Moreover, a torque of approximately 100 µNm (10-g force generation by 1 mm radius) is obtained at higher voltages – this torque value is the largest in 2–3 mm scale motors reported to date. Two micro-flat ultrasonic motors drive a miniature two-link as direct-drive motors without gears. The motion of the two-link is demonstrated by an open-loop burst wave control, which is a pulse-width modulation control for an alternative voltage. Although the demonstration is simple, it shows the feasibility of the smallest and most practical link mechanism. KEY WORDS M i c ro m o to r ; pi e zo e l e c t r i c ac tuator ; miniature link age; micro manipulato r Multiobjective evolution of deep learning parameters for robot manipulator object recognition and grasping Delowar Hossain and Genci Capi Assistive Robotics Laborator y, Depar tment of M echanical Engineering, Facult y of S cience and Engineering, HOSEI Universit y, Tok yo, Japan ABSTR AC T Deep Learning (DL) is currently very popular because of its similarity to the hierarchical architecture of human brain with multiple levels of abstraction. DL has many parameters that influence the network performance. In this paper, we introduce a multiobjective evolutionary algorithm (MOEA) to optimize the DBNN parameters subject to the error rate and the network training time as two conflicting objectives. To verify the effectiveness, the proposed method is applied to the robot object recognition and grasping task. We compare the performance of the optimized DBNN model with a) DBNN with arbitrarily selected parameters and b) Deep Belief Network-Deep Neural Network (DBNDNN). The results show that optimized DL has a superior performance in terms of training time and recognition success rate. In addition, the optimized DBNN model is effective for real-time robotic implementations. KEY WORDS D e e p l e a r n i n g ; m u l t i o b j e c t i v e e vo l u t i o n ; o b j e c t re co gni t i o n ; ro b o t grasping; DBNN Viewpoint optimization for aiding grasp synthesis algorithms using r e i n f o r c e m e n t l e a r n i n g B. Calli a, W. Caarlsb, M. Wissec and P. Jonkerc aRobotics Engineering Program, Computer Science Department, Worcester Polytechnic Institute, Worcester, MA, USA; bDepartment of Electrical Engineering, Pontifical Catholic University of Rio de Janeiro, Gávea, Rio de Janeiro, Brazil; cBioMechanical Engineering Department, Delft University of Technology, Delft, CD, Netherlands ABSTR AC T Grasp synthesis for unknown objects is a challenging problem as the algorithms are expected to cope with missing object shape information. This missing information is a function of the vision sensor viewpoint. The majority of the grasp synthesis algorithms in literature synthesize a grasp by using one single image of the target object and making assumptions on the missing shape information. On the contrary, this paper proposes the use of robot’s depth sensor actively: we propose an active vision methodology that optimizes the viewpoint of the sensor for increasing the quality of the synthesized grasp over time. By this way, we aim to relax the assumptions on the sensor’s viewpoint and boost the success rates of the grasp synthesis algorithms. A reinforcement learning technique is employed to obtain a viewpoint optimization policy, and a training process and automated training data generation procedure are presented. The methodology is applied to a simple force-moment balance-based grasp synthesis algorithm, and a thousand simulations with five objects are conducted with random initial poses in which the grasp synthesis algorithm was not able to obtain a good grasp with the initial viewpoint. In 94% of these cases, the policy achieved to find a successful grasp. KEY WORDS Active vision; reinforcement learning; robotic grasping; viewpoint optimization