Self-Driving Vehicle Architectures

This chapter describes the capabilities of self-navigating vehicle-processing architectures. It begins with a discussion of the types of sensors and data a self-driving vehicle receives as input while driving. We continue with a discussion of the requirements for ensuring the safety of the autonomous vehicle and its occupants, as well as for other vehicles, pedestrians, and stationary objects. Next is a description of the types of processing required for effective vehicle control. The chapter concludes with an overview of an example self-driving computer architecture.

After completing this chapter, you will have learned the basics of the computing architectures used by self-driving vehicles and will understand the types of sensors used by these vehicles. You will be able to describe the types of processing required by self-driving vehicles and will understand the safety issues associated with self-driving vehicles.

The following topics will...

Technical requirements

The files for this chapter, including answers to the exercises, are available at https://github.com/PacktPublishing/Modern-Computer-Architecture-and-Organization-Second-Edition.

Overview of self-driving vehicles

Several major motor vehicle manufacturers and technology companies are actively pursuing the development and sale of fully self-driving, or autonomous, motor vehicles. The utopian vision of safe, entirely self-driving vehicles beckons us to a future in which commuters are free to relax, read, or even sleep while in transit and the likelihood of being involved in a serious traffic accident is drastically reduced from the hazardous situation of today.

While this is the dream, the current state of self-driving vehicles remains far from this goal. Experts in the field predict it will take decades to fully develop and deploy the technology to support the widespread use of fully autonomous transportation.

To understand the requirements of autonomous driving systems, we begin with the inputs a human driver provides to control the operation of a current-generation motor vehicle. These are:

• Gear selection: For simplicity, we assume the presence...

Safety concerns of self-driving vehicles

At all times when some level of autonomous control is active in a moving motor vehicle, the algorithms behind the autonomous behavior must continuously apply a hierarchical set of requirements to meet the needs of passengers while making every effort to avoid negative outcomes, such as collisions with other objects.

The highest priority encoded in autonomous vehicle algorithms must always be to ensure the safety of the vehicle, its occupants, and others in the vicinity. Consider the alternative: if the vehicle’s highest priority was to get its passengers to the requested destination, the vehicle would interpret this as a license to run through red lights and strike pedestrians if those actions result in the quickest path to the destination.

The vehicle not only needs to predict and manage its path through the numerous obstacles that present themselves; it must also ensure that all of its safety-critical components are operating...

Hardware and software requirements for self-driving vehicles

Human drivers must sense the state of their vehicles and constantly evaluate the surrounding environment, keeping track of stationary and moving obstacles. The primary means of gathering this information is through vision.

Using eyesight, a competent driver monitors vehicle instrumentation, principally the speedometer, and scans the surrounding environment to perform lane keeping, maintain appropriate spacing from other vehicles, obey traffic signs and signals, and avoid any obstacles on or near the road surface.

Human drivers rely on other senses to a lesser degree, including the use of hearing to detect signals such as car horns and railway crossings. The sense of touch comes into play as well, for example when bump strips are installed on a highway surface to warn of an upcoming intersection. The sense of touch can also assist when an inattentive driver drifts off the roadway and onto the shoulder, which typically...

Autonomous vehicle computing architecture

Figure 16.2 summarizes the hardware components and the processing stages in an autonomous driving system based on the current state of technology, as described in this chapter.

Figure 16.2: Components and processes of an autonomous driving system

We introduced sensor technologies that gather information about the state of the vehicle and its surroundings. This information flows into a Sensing process, which receives data from the sensors, validates that each sensor is performing properly, and prepares the data for perception. The process of Perceiving takes raw sensor data and extracts useful information from it, such as identifying objects in video images and determining their location and velocity. With an accurate understanding of the vehicle’s state and all relevant surrounding objects, the Deciding process performs high-level navigation functions, like selecting which route to take to the destination, as well as low...

Summary

This chapter presented the capabilities required in self-navigating vehicle-processing architectures. It began by introducing driving autonomy levels and the requirements for ensuring the safety of the autonomous vehicle and its occupants, as well as the safety of other vehicles, pedestrians, and stationary objects. We continued with a discussion of the types of sensors and data a self-driving vehicle receives as input while driving. Next, we discussed the types of processing required for vehicle control. We ended with an overview of the Tesla HW3 computer architecture.

Having completed this chapter, you have learned the basics of the computing architectures used by self-driving vehicles and understand the types of sensors used by self-driving vehicles. You can describe the types of processing required by self-driving vehicles and understand the safety issues associated with self-driving vehicles.

In the next and final chapter, we will develop a view of the road ahead...

Exercises

1. If you do not already have Python installed on your computer, visit https://www.python.org/downloads/ and install the current version. Ensure Python is in your search path by typing python –-version at a system command prompt. You should receive a response similar to Python 3.10.3.

   Install TensorFlow (an open source platform for machine learning) with the command (also at the system command prompt) pip install tensorflow. You may need to use the Run as administrator option when opening the command prompt to get a successful installation. Install Matplotlib (a library for visualizing data) with the command pip install matplotlib.

2. Create a program using the TensorFlow library that loads the CIFAR-10 dataset and displays a subset of the images along with the label associated with each image. This dataset is a product of the Canadian Institute for Advanced Research (CIFAR) and contains 60,000 images, each consisting of 32x32 RGB pixels. The images...