I am not an engineer. Luckily, I talked to two very knowledgeable people: Brian Jee, an autonomous vehicle technical program manager and Chris Schwarz, Research Engineer at the National Advanced Driving Simulator. They gave me a lot of great information on the engineering/technological side of this new technology to make sure my book The Future is Autonomous: The U.S. and China Race to Develop the Driverless Car covered at least the basics of the automated driving system which drives AVs.
This article describes the basics of automation and autonomous vehicle “levels.” This is important because there is no fully autonomous vehicle on the market. Vehicle lie Tesla, despite the misleading name of Autopilot, still require a human to pay close attention to the road and take control of the vehicle if traffic conditions worsen, there is inclement weather, or in case of an emergency to avoid an accident.
Several basic concepts are helpful to get a better understanding of autonomy in general. Autonomy means an AI device does not need help from people to complete a task or set of tasks. Autonomous vehicles demonstrate autonomy to a degree depending on their level of automation. The different levels of automation are determined by standards created by the Society for Automotive Engineering (SAE). These standards are used by the National Highway Traffic Safety Administration (NHTSA). NHTSA has broad authority over vehicle safety and creates the Federal Motor Vehicle Safety Standards (FMVSS) in the US.
Many people already own a level one automation vehicle if it was purchased in the last ten years. Level one vehicles have very basic autonomous functions such as Anti-lock Braking Systems (ABS) and adaptive cruise control features that can accelerate or brake when driving.
Level two vehicles have things like steering and lane-keep assist, and the vehicle itself can drive. However, under level two the vehicle does not monitor the road and the driver is responsible for paying attention to the road and taking control. This is what Tesla vehicles are officially ranked while driving with Autopilot engaged. If autonomous trucks are driving in a platoon and closely following a human-driven truck, then that would also be level two.
At level three, the vehicle itself monitors the road and controls the braking, acceleration, and all other driving functions. Traffic jam assist would also be considered level three. The driver does not need to monitor the road. The vehicle will issue a “takeover warning” to the driver to take control. At that point, it would be the driver’s responsibility if there is an accident.
At level four and five autonomy, the vehicle controls all of the driving functions by itself. This including accelerating, braking, signaling and changing lanes, turning, and responding to hazards on the road. These hazards could include avoiding pedestrians as they cross the road, other vehicles, or other objects. At level four autonomy, the vehicle must already have a high-definition static map of the roads. The map is located in the cloud, which allows it to be updated remotely. At level five, the vehicle drives autonomously at all times, in all places, in any weather conditions, and does not require human intervention.
Brian Jee, an autonomous vehicle technical program manager, confirmed, “Mapping is just brutal.” He continued by saying a vehicle, with all of the cameras and sensors, would have to frequently drive down roads in a planned and coordinated process to ensure data quality and redundancy. Onboard cameras and sensors record millimeter-accurate data of everything on or near the road from the car’s perspective. The resulting map is called the delta, a high-definition (HD) map or, as engineers refer to them, semantic and geometric maps.
The mapping process is repeated multiple times through each lane, in addition to each direction of traffic. Not only does this take a long time, but it is also very expensive. It can cost $2,000 per linear kilometer. The maps also need to be constantly updated because the world changes. Someone could have graffitied a sign or there could be a new construction project. The difference between the real world and the mapped world is known as a delta. The expensive mapping process needs to be repeated every time a delta or map data issue is identified.
Because these maps must be constantly updated, how exactly will these updates be downloaded to the vehicle’s cloud computing system? One option would be through connection to a home WiFi network. However, another option would be through connection to a wireless network. Brian Jee described the importance of a Fifth Generation (5G) wireless network connection for autonomous vehicles. He said a 5G network assists autonomous vehicles in several ways. He said, “5G is the access (of the vehicle) to the rest of the world.” Because the different systems are computing at the edge of the internet or edge of what computer processing will allow, it needs a connection back to the server architecture. Therefore, a 5G network unlocks a lot of bottlenecks for the vehicle to drive safely.
For instance, Mr. Jee discussed how Tesla pioneered the Over-the-Air (OTA) updates system. When you’re at home, the car can connect to WiFi and download the update. According to Mr. Jee, with autonomous vehicles, “These updates will be more often and more significant. There will be huge changes to the deep learning models.” Continuing with the example of Tesla, Mr. Jee admitted the system had trouble recognizing traffic cones. With a 5G network connection, the vehicle could use the OTA updates system to more rapidly download a perception update so your car could recognize traffic cones and drive more safely.
The greater speed and capacity of the network would also allow for someone in an office to monitor the state of the vehicles remotely. This would be done either by autonomous vehicle companies themselves or they could license this task to a third party. Mr. Jee said, “If there’s a dispatcher monitoring how the car’s doing, monitoring any errors, or when the car gets stuck and doesn’t know what to do and it just stops in the middle of an intersection, then you’re going to be in trouble.” While this scenario sounds frightening, he assured me by saying that a 5G network connection would allow for a more robust and lower latency tele-operations person to recognize the problem and a remote technician to fix the problem and prevent a potential accident.
Specific details about how autonomous vehicles and their cameras and sensors work can be very complex. Hopefully this description about the different levels of automation and the mapping process clarifies, at least in part, how autonomous vehicles can drive themselves, at least at levels 3 to 5. I hope also that wit greater understanding will also come greater trust in AVs and, in time, acceptance of them as a safer vehicle transportation alternative to our current conventional driving system.
I will go into more detail about specific aspects of the technological process by which AVs drive themselves in future articles. However, if you like this description and would like to learn more about all of the potential benefits that AVs could provide consider buying my book on Amazon!
Use the link below to order your copy of The Future is Autonomous: The U.S. and China Race to Develop the Driverless Car today! For this month only the Kindle eBook versin is ONLY 99 cents! Get your copy before th deal runs out!
Also, if you like it, please don’t forget to rate and review the book on Amazon!