Weisong Shi - Computing Systems for Autonomous Driving

This Springer imprint is published by the registered company Springer Nature Switzerland AG

The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland

In the last 5 years, with the vast improvements in computing technologies, e.g., sensors, computer vision, machine learning, and hardware acceleration, and the wide deployment of communication mechanisms, e.g., dedicated short-range communications (DSRC), cellular vehicle-to-everything (C-V2X), and 5G, autonomous driving techniques have attracted massive attention from both the academic and automotive communities.

To achieve the vision of autonomous driving, determining how to make the vehicle understand the environment correctly and make safe controls in real-time is the essential task. Rich sensors including camera, LiDAR (light detection and ranging), radar, inertial measurement unit (IMU), global navigation satellite system (GNSS), and sonar, as well as powerful computation devices, are installed on the vehicle. This design makes autonomous driving a real powerful computer on wheels. In addition to hardware, the rapid development of deep learning algorithms in object/lane detection, simultaneous localization and mapping (SLAM), and vehicle control also promotes the real deployment and prototyping of autonomous vehicles. The autonomous vehicles computing systems are defined to cover everything (excluding the vehicles mechanical parts), including sensors, computation, communication, storage, power management, and full-stack software. Plenty of algorithms and systems are designed to process sensor data and make a reliable decision in real-time.

However, news of fatalities caused by early developed autonomous vehicles (AVs) arises from time to time. Until August 2020, five self-driving car fatalities happened for level 2 autonomous driving: four of them from Tesla and one from Uber. All four incidents associated with Tesla are due to perception failure, while Ubers incident happened because of the failure to predict human behavior. Another fact to pay attention to is that currently, the field-testing of level 2 autonomous driving vehicles mostly happens in places with good weather and light traffic conditions like Arizona and Florida. The real traffic environment is too complicated for the current autonomous driving systems to understand and handle easily. The objectives of level 4 and level 5 autonomous driving require colossal improvement of the computing systems for autonomous vehicles.

This book intends to present state-of-the-art computing systems for autonomous driving and to grab the attention of researchers and practitioners from both automotive industry and computer science and engineering community. The book consists of nine chapters, presenting the landscape, computing frameworks, algorithm deployment optimizations, systems runtime optimizations, dataset and benchmarking, simulators, hardware platforms, smart infrastructures, and open challenges for achieving L4/L5 autonomous driving vehicles, respectively. This book can be used by senior undergraduate students and graduate students in engineering and computer science majors. We hope this book will serve as a reference and a starting point for those who are interested in working in this field.

This book is a collective wisdom of the work from the Connected and Autonomous Driving Laboratory (CAR) at Wayne State University. We would like to thank all the past and current members in the CAR lab, including Sidi Lu, Qingyang Zhang, Yifan Wang, Xingzhou Zhang, Baofu Wu, Prabhjot Kaur, Samira Taghavi, Ren Zhong, Yongtao Yao, Ruijun Wang, Zhaofeng Tian, and Raef Abdallah. All of them contributed part of the content that is included in this book. We also thank our partners and sponsors who contributed hardware, software, and dataset and made these studies possible, including CalmCar, Continental, DENSO, Hesai, iSmartWays, Intel, Navya, Nvidia, PerceptIn, Toyota InfoTech, Velodyne LiDAR, Xilinx, and the City of Detroit.

As an essential part of the whole autonomous driving vehicle, the computing system plays a significant role in the whole pipeline of driving autonomously. There are two types of designs for computing systems on autonomous vehicles: modular-based and end-to-end based.

Modular design decouples the localization, perception, control, etc. as separate modules and makes it possible for people with different backgrounds to work together []. The main differences for these AV prototypes are the software and the configuration of sensors like camera, LiDAR, Radar, etc.