Inertial measurement unit (IMU) is a critical component for autonomous robot navigation. This paper presents a novel training framework that uses knowledge distillation to enhance low end, or low precision, IMU performance by transferring learned representations from high end IMU. In the proposed framework consisting of dual branch, teacher branch trained with high end IMU sensor data is used to train the student branch with low end IMU. For training the low end IMU branch, combined loss function that balances signal matching and knowledge transfer is used. Testing with the MAGF-ID dataset demonstrates substantial reduction of mean absolute error as much as 39-65% with accelerometer measurements and 47-70% for gyroscope readings with best improvement in the (gyroscope) X-axis, as compared to original low end IMU measurements. Qualitative analysis also reveals that proposed framework is efficient in systematic error correction and noise suppression.

We describe an approach to automation of visual inspection of ball grid array (BGA) solder joint defects of surface mounted components on printed circuit boards by using a neural network. Inherently, the BGA solder joints are located below its own package body, and this induces a difficulty in taking a good image of the solder joints when using a conventional imaging system. To acquire the cross-sectional image of a BGA solder joint, an X-ray cross-sectional imaging method such as laminography and digital tomosynthesis is utilized. However an X-ray cross-sectional image of a BGA solder joint, using laminography or DT methods, has inherent blurring effect and artifact. This problem has been a major obstacle to extracting suitable features for classification. To solve this problem, a neural network based classification method is proposed. The performance of the proposed approach is tested on numerous samples of printed circuit boards and compared with that of a human inspector. Experimental results reveal that the proposed method shows practical usefulness in BGA solder joint inspection.

Automation of welding process in shipyard is ultimately necessary, since welding site is spatially enclosed by floors and girders, and therefore welding operators are exposed to hostile working conditions. To solve this problem, recently a robotic welding system for closed block assembly has been developed. For complete autonomy, this system needs a sensor system for the weld seam tracking and the recognition of the working environments. In this paper, a three-dimensional laser vision system is developed based on the optical triangulation technology in order to provide robots with work environmental map. This sensor system composed of three laser stripes and two visions is used to detect the welding positions and to recognize the 3D shape of the welding environments. The algorithm architecture for the welding environment recognition is divided into two parts. The first is the conventional 3D scanning module, and the second the plane generation module utilizing a Hough transform. The design of the sensor system and the algorithm for sensing the structured environment are described and discussed in detail.

In this paper, we described an approach to automation of visual inspection of solder joint defects of SMC (surface mounted components) on PCB (printed circuit board) by using neural network and fuzzy rule-based classification method. Inherently, the surface of the solder joints is curved, tiny and specular reflective; it induces a difficulty of taking good image of the solder joints. The shape of the solder joints tends to vary greatly with soldering condition and the shapes are not identical to each other, even for solder joints belonging to a set of the same soldering quality. This problem makes it difficult to classify the solder joints according to their qualities. Neural network and fuzzy rule-based classification method is proposed to efficiently make human-like classification criteria of the solder joint shapes. The performance of the proposed approach is tested on numerous samples of commercial computer PCB boards and compared with the human inspector's performance.

In this paper, we present the design and the structure of our multiple micro-robot system, CENDORI, designed for playing a robot soccer game, MIROSOT. A robot soccer game is a very complex robot application that incorporates real-time vision, robot control wireless communication and management of multiple robots. To build a robot system with high performance and reliability, we designed the system architecture with a centralized concept. The robot control relies largely upon the vision system and the host computer generating low level motion control words. The experimental results and the performance of the robot are discussed.