SBIR Phase II: Versatile Robot Hands for Warehouse Automation

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this project affects one of the fastest-growing sectors of the US economy. E-commerce sales in 2015 accounted for 7.4% of total U.S. retail and are expected to rapidly rise. The potential for the commercial impact of general each-picking systems is high, as current manual labor methods are pain points for distribution centers; human picking is unpleasant, expensive and inefficient due to high absenteeism, high turnover and human error. The success of the proposed technology will also contribute to American competitiveness in the robotics industry. Of the top 20 distribution system integrators, only three are currently based in the U.S. Robotics is going to be the key driver of progress in this area, where each-picking, our core product capability, is a key component of future automated distribution systems. Beyond warehousing logistics, applications that our technology can benefit include: broad applications of industrial automation and manufacturing; military applications (e.g., IED disposal, where robots can perform tasks that are dangerous for humans to perform); and assistive healthcare (e.g., where robots must be compliant enough to be safe around humans while interacting successfully with unknown environments).This Small Business Innovation Research Phase II project will focus on the development of a state-of-the-art each-picking robotic system and its deployment, initially targeted at the order fulfillment industry. To date, robotic systems have enabled significant progress on transporting inventory on shelves or in totes. However, there has not yet been a deployed system that can perform the task of picking individual items from inventory bins and placing them in boxes for shipment. During Phase I of this project, RightHand Robotics developed a picking system far in advance of the research literature on robotic grasping, picking tens of thousands of items previously unseen objects, with error rates of less than 0.1%. During Phase II, the project will focus on advancing the state of the art in data-driven refinement of grasp planning using machine learning techniques, and will develop methods for box-packing that exploit the company?s advanced compliant grippers. These improvements will result in an average pick-and-place time of 6 seconds or less and an undetected placement failure rate of fewer one in ten thousand.