Psychology, not technology, appears to be the main barrier toward the adoption of autonomous vehicles. A survey by the American Automobile Association (AAA) released in May 2018, found that “73 percent of American drivers report they would be too afraid to ride in a fully self-driving vehicle, up significantly from 63 percent in late 2017.”
In the same way airplane manufacturers and airlines follow a stringent regimen of testing and transparency through design, manufacturing, and operations to gain their constituents’ trust, the automotive industry can rely on innovative and thorough verification and calibration to accelerate and secure the trust of its own stakeholders.
Lessons from the Airline Industry
When considering the main phases of a flight—taxi, takeoff, climbing, cruising at altitude, descent, approach, and landing—autopilot is used for climbing, cruising, and descent, making most of the flight time without direct pilot intervention. The pilot enters the flight plan in the flight management system. The planes’ computerized system then relies on a sophisticated network of sensors to continually assess and adjust the speed, attitude, direction, and other factors.
Let’s compare safety records between human-driven motor vehicles, and highly tested and automatized air transportation. For the road case, the National Highway Traffic Safety Administration reported over 35,000 motor vehicle fatalities in 2015, 1.13 fatalities per 100 million vehicle miles traveled, and nearly 11 fatalities per 100,000 U.S. residents.
For the air case, U.S. airline accidents (planes with more than 10 passengers) recorded by the National Transportation Safety Board revealed no fatalities in 2015. Of the accidents that did occur, just 0.155 happened for every 100,000 flight hours and a rate of 0.0035 per one million miles flown.
According to the National Safety Council, Americans have a one in 114 odds of dying in a car crash and one in 9,821 odds of dying in air and space transport incidents. Quite a difference!
The air industry is known to meticulously test, maintain, and inspect their equipment. It also relies on a pervasive and effective vehicle to infrastructure communication framework made of traffic control and guiding systems. How many motor vehicle drivers go through a checklist before driving off, and how many cars communicate with the infrastructure today?
Although it is clearly more dangerous to be in a car than in an airplane, when an airplane goes down it makes the news in a sensational manner—and that’s the conundrum autonomous vehicles are facing today.
The Journey to Autonomous Driving
Human drivers are self-confident and were in control while on the road. Many drivers even look down at automatic transmission as reducing the scope of control afforded to them. However, human reflexes are far outpaced by sensors and computerized systems that allow autonomous cars to react earlier, faster, and more precisely than drivers can.
The Society of Automotive Engineers and the National Highway Traffic Safety Administration define six levels of autonomous driving (Figure 1), ranging from Level 0, where the human driver controls everything, to Level 5, where the vehicle’s capabilities are equal to (or even better than) a human driver in every scenario. A Level 5 autonomous vehicle requires no human intervention (or even interaction) in any scenario.
While artificial intelligence and algorithms are often the stars of articles on autonomous vehicles, they would literally be blind, deaf, and disoriented without the variety and quantity of input provided by a diverse set of sensors. Anyone who has played the Marco Polo or piñata games learned that a lack of proper input dramatically limits one’s ability to realize their potential.
Each sensor performs a role and sometimes overlaps other sensors to mitigate any shortcoming by aggregating the most complete situational information at any time and in any circumstance (Figure 2).
- Global Positioning System (GPS) provide data for calculating current position and available routes to the final destination. They work in most conditions, except when occulted from satellite line of sight, in conditions like heavy forests, dense high-rise urban centers, or covered structures like tunnels, buildings, or large bridges.
- Cameras help judge distance to objects, avoid obstacles, and detect and analyze 2D information like lane markers and road signs. While inexpensive by themselves, they require costly image processing systems to interpret the captured images. They also struggle in situations of limited visibility such as dark, rainy, snowy, or foggy conditions.
- Light Detection and Ranging (lidar) sensors provide a detailed 3D representation of terrain and objects in the surrounding environment. These sensors offer high-definition images but are still very expensive and have not yet been proven in mass-market applications. They also rely on moving parts for rotating 360-degree scans and generate massive volume of data that require tremendous signal processing power and data management sub-systems. As they depend on light, they suffer from the same environmental limitations as cameras.
- Short-range radars detect objects in the immediate proximity around the car. They have been proven in mainstream deployments but do not offer the fine image definition that cameras or lidar can provide.
- Longer range radars aid with distance and speed control, brake assist, and emergency braking. They are especially reliable for all-weather autonomous driving. While legacy 24 GHz radar sensors were bulky and lacking in resolution, new high frequency, high bandwidth, smaller 76-81 GHz radar sensors can differentiate objects only a couple of inches apart. They, however, cannot offer the fine image definition that cameras or lidar can provide.
Of course, sensors are only as good as their precision and resistance to disturbance. A sensor, either badly calibrated or perturbated by other nearby equipment would, provide incorrect information to the central decision-making system.
To demonstrate the potential for seeing zero fatal car accidents, autonomous vehicle developers need to optimize their design and test methods, and prove the reliability and safety of their solutions without any doubts. Their need to minimize loss, noise, and frequency response errors demands sophisticated, reliable, and precise test solutions that ensure repeatability of the test results. Given the accelerating rate and diversity of technologies involved in autonomous vehicle platforms, a strong collaborative partnership is required between component manufacturers, automobile makers, and testing solutions providers.
Communication across the Board
Although the word “autonomous” in autonomous vehicles is evoking an image of completely independent platform, communications capabilities significantly enhance their ability to be informed to anticipate challenges, while also sharing information to help others.
The modern autonomous vehicle will communicate broadly across a variety of frameworks (Figure 3): vehicle-to-vehicle (V2V), vehicle-to-network (V2N), vehicle-to-infrastructure (V2I), vehicle-to-pedestrian (V2P), vehicle-to-utility (V2U), and eventually vehicle-to-everything (V2X).
Automotive designers can choose between two major categories of wireless communications technologies, with a third one coming fast with the potential to overtake the currently available ones.
- Dedicated Short Range Communications (DSRC) is built on the IEEE 802.11p standard, derived from the proven WiFi 802.11a technology. The two key benefits of 802.11p DSRC are its availability for the automotive industry and a very low latency around 5 ms. Automakers who want to deploy V2X communications right now can do so immediately. However, it requires the installation of many new access points and gateways along the roads, increasing time and cost of full deployment.
- 4G-cellular LTE uses the existing cellular network infrastructures, providing better security and longer communication range. However, current 4G LTE networks do not provide the low latency needed to enable critical V2V emergency communications as it varies between 30 ms and 100 ms.
- 5G-cellular will most likely replace both DSRC and 4G C-V2X (LTE-V) as neither of them meet the stringent requirements of mission-critical autonomous driving and ADAS systems. Once finalized and deployed, 5G will deliver the concrete benefits of 20 Gbps data rates, up to 500 km/h mobility, less than 1 ms latency, high-density of connections, and ultra-reliability required by autonomous driving cars and ADAS. On the downside, it is not fully formulated yet and it will take some time for wireless service providers to fully deploy 5G across their entire infrastructure.
Bringing It All Together With Networking
Connecting all automotive systems, the wiring harness is now the third heaviest component in a vehicle, as well as its third most costly system. The wiring harness is also responsible for 50 percent of labor costs during automobile assembly. Autonomous vehicles also need a high-bandwidth, low latency network to connect all sensors, cameras, diagnostic, communications, and central artificial intelligence.
The emerging solution is Automotive Ethernet (Figure 4). In the same way that WiFi is the foundation for DSRC, Ethernet is a well-known, well-trusted, and ubiquitous solution in traditional local area networking (LAN). The advantages of Ethernet—multi-point connections, higher bandwidth, and low latency—are highly attractive to manufacturers. However, traditional Ethernet is simply too noisy and interference-sensitive for use as-is in automobiles. The IEEE has defined the physical layer under the 100Base-T1 standard and the necessary protocols of time synchronization (IEEE 802.1AS), audio video bridging transport (802.1Qav), scheduled traffic transmission (IEEE 802.1Qbv), and TCP/IP stack.
Automotive Ethernet is the right choice for the autonomous driving and the ADAS systems. However, Automotive Ethernet is very new to the automotive industry and requires extensive testing of the transmitter, receiver, link segment, and higher layer protocol functions.
Testing Can Accelerate Adoption and Help Save Lives
When the biggest challenge is trust, testing and sharing results with the public during design, manufacturing, and operations of this new category of vehicles will be central to build the confidence required.
Every delay in adopting autonomous vehicles brings another day that leads to people losing their lives to human drivers. Testing contributes to breaking the psychological barrier and can accelerate the adoption of autonomous vehicles. Lives will be saved as automotive, communications, and testing leaders collaborate to rapidly and safely bring trusted and innovative transportation solutions to market.