SBIR Phase I: Autonomous harvesting, mapping, and forecasting for fresh produce through application of robotics, computer vision, and machine learning

点击翻译按钮获取中文摘要

The broader impact/commercial potential of this project is significant. Agriculture is one of the most important industries impacting the economy, society, and the environment. By automating harvesting of fresh produce, society's ability to grow healthy and sustainable food will increase substantially. The innovation will enhance scientific and technological understanding of how to deploy commercially viable multi-robot coordinated groups performing complex automated tasks, in agriculture and beyond. The long-term opportunity for agricultural robotics is $150B+. This innovation focuses on controlled-environment methods, which are more intensive in terms of capital and labor, but far less intensive in terms of water, chemical use, and fertilizer use. If robots can reduce manual labor of large-scale production, and enhance human management precision, it will enable intensive agriculture practices to compete with, and in some cases supplant, traditional chemical-, labor-, and water-intensive approaches. This will make the US a more competitive global producer of high-quality produce, increase food security, increase access to healthy food for all people, and protect natural resources and the environment.This Small Business Innovation Research (SBIR) Phase I project will advance the fields of Robotic Applications, including computer vision/machine learning and robotics controls, by solving critical problems faced when operating in highly dynamic yet precise biological environments like farms. 1) By evolving and combining approaches from the forefront of computer vision, the project will improve environment sensing of e.g. clusters of fruit, their locations, conditions, precise locations of stems and obstacles. 2) Using that information, the project will develop innovative approaches to complex planning problems where leaves, stems, and obstacles interfere with optimal harvesting but are movable. This will be done with millimeter-level precision, where current autonomous planning approaches operate on multi-centimeter precision. 3) Lastly, the project will combine environmental data with 3D reconstructions of grow operations and individual fruit tracking in order to provide better sampling data to forecast production on a 1-4-week basis, and eventually build a deep learning model. This project is critical to achieving the commercial threshold of performance for robotic harvesting services, including high percent of berries picked, acceptable speed of picking, and the provision of climate sensing and forecast information.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.