A Dual-driven Soft Robotic Hand Based on Hydraulic System and Strings
Author: Ding, Zhiyuan
Abstract: Soft robotic hands have the benefits of high human-interaction safety and the ability to grip irregular objects compared to rigid robotic hands. However, it is difficult to achieve high gripping force and precise manipulation. In this soft robotic hand project, a dual-driven system composed of a hydraulic system and an inelastic string system is applied to improve the soft robotic hand and generate higher gripping force. The string system originates from biological tendons. These systems form the basic modular FSEA (fluid and string driven elastic actuator) units, which assemble on the base framework to create the soft robotic hand. The designs of the hydraulic system are first explored, using qualitative observations to advance from using balloons to plastic sealing bags. Then the focus shifts to the design of the silicone fingers, in which the manufacturing process and geometry are experimented to achieve the best performance. Based on the structure, a controlling system based on the MCU board ESP32 and a set of sliders to remotely control the robotic hand via Bluetooth are programmed. Finally, the performance of the robotic hand is demonstrated through quantitative experiments on its bending angle and gripping force, and it is compared with existing soft robotic hand projects. Its real-life applications are shown through gripping objects of various sizes and materials and forming different gestures. Overall, a soft robotic hand that has a comparably high gripping force of approximately 4N per finger and a maximum turning angle of around 140° is built from modular FSEA units on an extendable platform, showing that the dual-driven system and the use of liquid in the soft robotic hand can elevate its performance significantly.
A Dual-driven Soft Robotic Hand Based on Hydraulic System and Strings.pdf
Hand Gesture Recognition Under Low-Light Conditions
Author: Alex He
Abstract: Human gesture recognition has become more and more important in recent years. However, issues such as image lighting interfere with gesture recognition. In this paper, a 3-part model involving image enhancement, keypoint extraction, and gesture recognition is proposed to alleviate the issue of insufficient lighting, and the results are given.
Hand Gesture Recognition Under Low-Light Conditions.pdf
基于水下滑翔机机翼截面翼型的相关研究
Author: 赵耀楠
Abstract: The Underwater Glider (UG) is a new type of underwater exploration robot. Instead of using traditional external propulsion, it uses a water storage system at the front of the body to change its own buoyancy to ascend and descend, and a rudder to change the attitude during sinking and gliding to achieve the horizontal direction of rotation. Due to the low frequency adjustment interval of its own buoyancy and center of gravity, the underwater glider consumes minimal energy and is therefore suitable for long duration, wide range marine environmental monitoring tasks. Compared to similar instruments for ocean exploration, such as underwater exploration robots and ocean exploration vessels, underwater gliders are more energy efficient and can perform longer and wider range of ocean detection tasks.
The development of underwater glider technology first began in the United States. After the oceanographer Stommel proposed the development and application plan of underwater gliders in 1989, the U.S. launched Slocum, Seaglider, and Spray in the 1990s. These three underwater gliders have become the mainstream products in the international market after continuous research and improvement. The development of this technology started in China in the 21st century, and the research on underwater gliders was started by Tianjin University in 2002.
"In March 2017, the "Sea Wing-7000" of Shenyang Institute of Automation, Chinese Academy of Sciences completed the first large depth observation mission of 6329 meters in the Marianas Strait. In April 2018, the "Haiyan-10000" deep-sea underwater glider of Qingdao National Laboratory of Marine Science and Technology (part of Tianjin University) completed a dive of 8,213 meters in the Mariana Strait, again setting an international record for underwater glider operating depth. The world record for the working depth of deep-sea underwater glider has been set again.
The two iterations of underwater glider models studied in this paper uses simple mechanics to achieve changes in buoyancy and the center of gravity to simulate the working principle of the glider. In this paper, we will introduce the dynamic model of the glider through force analysis, introduce the preliminary design of the glider components, design the experiments to test the performance of the glider, adjust the design parameters of the second generation glider based on the preliminary test results, redesign the experiments, and compare the performance of the two generations of gliders.
基于 SLNet 的半监督手语识别多模态情感交互系统
Author: 宋卓卿
Chairman Xi emphasizes that the disabled is a specifically tough group that needs special care and attention. The disabled should integrate into the society, and it is worth to pay attention to the problem of how to reduce the obstacle between the disabled and the society. In addition, facing inconvenience in work and daily lives, the disabled will more easily arouse psychology problems than others, so the consolation to them is also a problem awaiting to be solved. Based on the background, this project, targeting at people who have difficulty in language, exploits a interaction system based on SLNet's semi-supervised sign language recognition and realizes functions like sign language recognition, condolation, and etc.
The main innovations of this project are as follows:
1. Propose a complete set of automatic sign language recognition data set recording system. This system can change and add new gestures according to user needs, record sign language data in various environments, and improve the recognition rate of sign language recognition system.
2. Semi supervised learning based on pseudo tags is introduced into emotional interaction system for the first time. Aiming at the heavy data annotation problem caused by supervised learning and improving the accuracy of subsequent online recognition, the introduction of semi supervised learning based on pseudo tags into emotional interaction system can greatly improve the accuracy and robustness of online recognition of the system.
3. Crawl and sort out the boast data set from Douban, and train the emotional appeasement robot based on seq2seq+attention framework. At the same time, in order to solve the lack of information richness in the traditional single attention mechanism, a multi tap combined attention mechanism for this chat robot is proposed to optimize the results.
Emotional chat robots can provide sign language communication services to people with language disabilities and emotional comfort at the same time, so that AI technology can give love and comfort this special social group.