Can do: the ongoing role of CAN networks in autonomous vehicles
There has been a steady increase in the demands on the computing ability of automobile communication networks as more and more technologies have been added to vehicle builds, from LED screens to smart cameras, and sensors to in-car Wi-Fi. But the development of autonomous vehicles (AVs) has brought about a sea of change.
AVs depend on robust computing systems to perform all the functions previously left to the vehicle operator. In that sense, the vehicle’s network needs to be able to make the same complex, split-second, life-saving decisions that a human brain can. The onboard infrastructure required to replace a human driver—including GPS, LiDAR, radar, and video—demands exceedingly more bandwidth than in previous vehicle networks.
With the advent of “disruptive” technologies such as Ethernet networks, it can be easy to assume that current automotive computing standards, such as the CAN bus, are on the way out. The reality of the situation is that CAN networks are still used in virtually all automobiles, including the most “cutting-edge” vehicles. This is exemplified by Nvidia—one of the industry’s leading AV chipmakers—employing the standard CAN protocol in their most innovative AV supercomputing platform.
Old technology vs. new technology
Ethernet and Flexray networks have emerged as the communication backbones of choice for supporting these types of high-speed data. And the numbers bear this out; both options are much more powerful than previous networks.
Flexray, developed in the early- to mid-2000s, is capable of 254 bytes at 10 Mbps, while an Ethernet frame is capable of 1500 bytes at 100 Mbps. A traditional CAN network maxes out at 8 bytes at 500 kbps, while the more powerful CAN FD still only offers up a 64-byte payload and 2-Mbps transmission speed.
This leaves a huge gap between old and new technologies. But, as Nvidia has recognized, the advent of new technologies doesn’t necessarily mean the end of the old.
Breathing new life into old networks
Nvidia is building its industry-leading Drive PX Pegasus computing modules with multiple types of inputs and outputs including CAN; Flexray; 16 dedicated high-speed sensor inputs for camera, radar, lidar and ultrasonics; as well as multiple 10-Gbit Ethernet connectors. With a combined memory bandwidth of over 1 Tbps, the Drive PX Pegasus is far and away the most powerful chip yet created for autonomous vehicles, as well as the company’s first Level 5 system.
Jesse Paliotto is Director of Marketing at Kvaser AB.
So, why the patchwork approach to platform architecture? Very simply, Nvidia knows that, along with CAN’s limited bandwidth, comes time-tested reliability. While Nvidia may be trying to recreate the car, it isn’t trying to recreate the wheel.
By maintaining CAN, the Drive PX Pegasus allows the less powerful network to focus on what it’s best at—handling less data-intensive but critical tasks such as powertrains, power steering, and anything requiring hard real-time control. And, as always, CAN provides data consistency, guaranteed latency, network-wide error detection, and fault confinement. It’s a case of simplicity perfectly complementing complexity to produce something elegantly suited to the task.
While most “disruptors” aim to create something radically new, Nvidia’s use of established technologies to complement its newer innovations has the feel of something radical.
It’s an endorsement of sorts, and the results speak for themselves. Currently, more than 25 companies are using Nvidia chips to develop Level 5 self-driving systems, including Tesla, Zoox, Optimus Ride, TuSimple, Yandex, and nuTonomy.
As computing demands continue to increase moving forward, it’s anyone’s guess what the long-term future holds for most of the networking technologies mentioned, but if the case of Tortoise v. Hare has taught us anything, it’s that slower and more reliable should never be counted out of the race.