CAREER: From Grasp Quality to Hand Quality: Analysis and Optimization for Effective Robot Hands

点击翻译按钮获取中文摘要

Robotic manipulators often exhibit high precision, repeatability, and grip strength, characteristics which have enabled many applications, primarily in industrial settings. However, robot hands still lack versatility: once a manipulation task implies variation, it quickly falls beyond the capabilities of today's robots. This prevents robot manipulators from operating in many environments, from cluttered store rooms and warehouses to typical homes and other human settings. This project aims to develop and demonstrate more versatile robot hands, able to manipulate a wide range of objects and to operate in clutter. Robots equipped with such hands could perform what are otherwise injury-prone tasks like kitting and bin picking (in manufacturing), or order picking and packing (in logistics). In healthcare, more versatile robotic manipulators could assist with activities of daily living, many of which involve contact and manipulation. From an educational perspective, this project aims to encourage interdisciplinary training by combining areas of robotics traditionally spanning multiple engineering disciplines, and to promote diversity in engineering education by inspiring and preparing young students for STEM careers, at both the high school and undergraduate levels.The key research objective of this project is to find new ways of building and using robot hands, where the high level grasp planning algorithms, the low-level control loops, and the hardware design itself are optimized simultaneously, in a tight loop, and taking each other into account. This represents a departure from the "build, then program" approach, where algorithms for performing complex tasks (such as grasping and manipulation) are researched only after the underlying hardware (the hand) has been designed and constructed. By combining these stages, we can jointly optimize algorithms, sensor arrays, control strategies and mechanism structures that fit and complement each other. Both mechanical and algorithmic complexity can then be increased based strictly on provable performance gains.