CAREER: Towards Fundamentals of Adaptive, Collaborative and Intelligent Radar Sensing and Perception

点击翻译按钮获取中文摘要

Automotive radar imaging represents a pressing technological need in perception for automotive active safety and autonomous driving. Automotive radar operating at millimeter-wave frequencies (typically around 77 GHz) is indispensable due to its superior capability in measuring range, velocity, and offering better perception performance in occlusion situations under all weather conditions and much lower cost than Lidar. The state-of-the-art automotive radar is prone to mutual interference and multipath issues, and its limited angular resolution—approximately 13 degrees—is inadequate for facilitating perception tasks for fully autonomous driving. This CAREER project aims to innovate the automotive radar perception in service of well-being of individuals in society and reduction of fatal accidents on U.S. highways by exploring adaptive, collaborative sensing and radar imaging physics principles. It deepens our understanding of how such sensing can enhance the dynamic range and address the ill-posed nature of the radar imaging inverse problem. Addressing the challenges in automotive radar imaging necessitates collaboration between academia and industry. Through collaboration with automotive industry sector, this project will make the research more impactful and pertinent, hastening the transition of research findings into automotive industry applications. Autonomous vehicle research helps attract underrepresented researchers. Through K-12 outreach activities and recruiting underrepresented groups and women in engineering, this project will promote diversity in the STEM research, inspiring them to pursue advanced research in radar field. This CAREER project addresses the ill-posed inverse problem inherent in automotive radar imaging by investigation of learning based adaptive and collaborative methodologies. The project tackles the robust radar perception for autonomous vehicles problem through incorporation of the radar imaging physics to drive the design of innovative machine learning algorithms. Intelligent signal processing and machine learning techniques will be developed at multiple layers, including 1) learning-based adaptive radar transmit parameter adjustment, 2) iterative optimization algorithms to enhance the dynamic range of automotive radar sensing by exploiting constructive interference, 3) model-based learning framework for collaborative high-resolution radar imaging in an automotive radar network, 4) physics-aware radar machine learning algorithms for robust environment perception. These new techniques will demonstrate how the science insights advance high resolution radar imaging, and robustly detect and classify objects in the highly dynamic autonomous driving environment.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.