Beyond Autonomous Cars

The ecosystem of semi-autonomous vehicles goes far beyond self-driving cars, and so do the design challenges.


As the automotive industry takes a more measured approach to self-driving cars and long-haul trucks for safety and security reasons, there is a renewed focus on other types of vehicles utilizing autonomous technology.

The list is long and growing. It now includes autonomous trains, helicopters, tractors, ships, submarines, drones, delivery robots, motorcycles, scooters, and bikes, all of which are in various stages of development.

Fig. 1: Autonomous tractors could change agriculture as we know it. Source: John Deere

Fig. 1: Autonomous tractors could change agriculture as we know it. Source: John Deere

Industry consensus says the bulk of the profit, as well as the life-saving potential, lies in developing self-driving consumer cars. However, tech executives say looking at the autonomous ecosystem holistically reveals important takeaways for the hardware community.

“The industry’s focus on autonomous cars is correct from a safety standpoint, given the number of cars being driven and the potential to dramatically decrease road deaths,” said Paul Karazuba, Expedera’s vice president of marketing. “But it also makes sense to look at things like tractors and ships and aircraft for certain levels of autonomy, especially if you can make it safer, more efficient, greener, or whatever the case may be.”

For one thing, autonomous automotive innovations find their way into other autonomous vehicles, and vice versa. And all of those innovations find their way into other industrial and commercial applications.

“Everybody talks a lot about autonomous cars, but it’s actually all modes of transportation that we’re chasing,” said Walt Hearn, vice president of the Ansys Americas sales team, in a podcast earlier this year. “We’re chasing autonomous boats, autonomous airplanes, autonomous trains, because all the sensors are able to be used for all different industries. It’s not just being pushed in automotive. Imagine a lidar and a camera sensor that are on an autonomous car. Now it’s in a manufacturing plant monitoring the robots that are manufacturing products. There are so many applications coming out of autonomy.”

That is partly due to the different financial assumptions for autonomous technology in vehicles other than cars. While keeping production costs as low as possible is a major concern for consumer cars, the calculation is different for other sorts of commercial vehicles. “There are less price, space, and power restrictions on commercial vehicles like trucks, ships, and aircraft,” said Robert Schweiger, group director of automotive solutions at Cadence. “Cost is also of less concern because of the return on investment of operating these vehicles 24/7 and saving labor costs.”

Many of the technical challenges are also easier in other vehicles. “Autonomy feeds on predictability, and really does not like unpredictability,” said Thierry Kouthon, technical product manager at Rambus. “The unpredictability tends to be more on the consumer side. Professional and business operations are more predictable, which is why it’s an interesting way to look at things because it narrows down the complexity of the problem. Autonomy is basically about managing sporadic evidence, and the less sporadic evidence you have to manage, the better the autonomy,” he said.

There are certain technical aspects to autonomy that tend to remain consistent, whether the vehicle is a car or a forklift. Kouton noted that autonomy in vehicles that are traditionally operated by humans means automating or electrifying all the functions that are usually either mechanical, hydraulic, or otherwise. “Instead of having a hand or foot pulling a lever or pressing a pedal, you have a computer that is engaging an actuator. You need a computerized environment to control all these functions, usually an ECU. All these ECUs have to work in harmony to deliver the autonomy, meaning that the brakes have to listen to the engine, which has to listen to the steering.”

Furthermore, AI and ML allow the vehicle to operate in a wide variety of circumstances. Network connectivity is crucial for a wide swath of autonomous vehicles, and is believed to be one of the keys to unlocking L5 fully-autonomous cars. Equally important is what happens when that connectivity suddenly disappears. Through the process known as graceful degradation, self-driving cars must be equipped to pull over to the side of the road and park in the case of connectivity issues. In the case of an air taxi, that means being able to withstand brief connection losses without falling out of the sky.

Autonomous technology is also inherently disruptive to the EDA design cycle, said Neil Hand, director of strategy for design verification technology at Siemens Digital Industries Software. “Autonomous vehicles, whether they be robots, cars, or planes, bring in a whole new set of requirements. It adds new functional safety aspects, or a new focus on non-determinism that has to be managed throughout the flow.”

Much of the disruption comes from the shift from an electromechanical focus to an electrical and software focus, which can be a major transition for vehicle companies that are used to a certain design protocol.

“Momentum and inertia have a huge impact on how the system is designed and what the compute requirements actually are,” said David Fritz, vice president of hybrid-physical and virtual systems for automotive and mil/aero at Siemens. For one thing, a maneuver that might be completely safe for an autonomous vehicle could be inappropriate and possibly lethal to any sort of autonomous vehicle that carries a human inside of it. “It’s designing a vehicle such that it’s cognizant of the fact that not only does the vehicle have momentum, but so do the passengers, and the anatomy of the passengers has something to do with whether or not a decision the vehicle could make is safe for the passengers.”

From a data perspective, autonomous vehicles generate far more data than their less-autonomous counterparts through myriad sensors, cameras, and other data-generating components. That data volume in turn impacts memory, bandwidth requirements, and data center usage.

“Data is no longer generated by human events,” wrote Mike Gianfagna, senior director of enterprise marketing at Synopsys. “Thanks to widespread sensor deployment, coupled with a hyperconnected environment, all types of devices are generating data at an exponentially increasing rate. Your smartwatch captures details about your exercise regimen and your health. According to one study, an autonomous vehicle can generate 5TB of data per hour of operation. If you consider how many such vehicles will be in operation in the coming years, you can clearly see a data avalanche.”

Because of this data explosion, back-end compute and data storage will be a major challenge, as well as a business opportunity for vehicles across the autonomous spectrum, which is more than just cars.

Safety and security
Some of the more glaring differences from a hardware perspective between autonomous cars and trucks compared to other autonomous vehicles are related to safety and security. A robo-taxi without the proper hardware and software is a potential killing machine, but lethality would be a challenge for an autonomous skateboard. More dangerous non-car autonomous vehicles are unlikely to reach the same levels of mass adoption as consumer automotive. They also are less likely to encounter ethical issues that require prioritizing human life. A self-driving car, for example, must consider the well-being of the people inside and outside of the vehicle.

Many autonomous vehicles outside of the consumer automotive industry are unmanned or require only infrequent or remote human participation, noted Frank Schirrmeister, vice president of solutions and business development at Arteris IP. “This significantly streamlines the necessary decision-making and AI the technology necessary to process and execute those decisions. In a car, the system has to realize that someone is likely to get hurt in the next 10 seconds, and then it has a choice between the different people involved. But in many other cases, a vehicle can just self-destruct, and that will save a life.”

But how a vehicle behaves depends on more than just the initial design. Rambus’ Kouthon said high-tech cars are by far the most popular target among commercially available semi-autonomous vehicles, because they are the most common. “Still, the strategy is often similar regardless of the vehicle type. Basically, a hacker can go in and make one ECU believe something is false. For instance, it will make the engine control unit believe that the engine is running at 20 miles per hour, when it is actually at 100 miles per hour, which then creates an accident. Or it can tell the brakes to slam on when the car is running at 80 miles per hour. Those are some of the actions you can take as an adversary if you can access the ECUs — both the unit itself and the communication between the units or the software that runs those units. Products based on cryptography and security engineering can provide some amount of protection.”

By land, sea, or air
When comparing the hardware considerations of autonomous tractors, ships, and aircraft to that of cars, tractors almost always will have a line of sight to the sky and thus be able to make use of GPS, which is not always the case with a car.

“It’s not uncommon in the Financial District in San Francisco to have your car think you’re a block or two away from where you actually are because of the height of the buildings,” Expedera’s Karazuba said. “On the other hand, tractors will encounter wildlife to a degree that most cars will not, will have to operate amid huge clouds of dirt and dust, and also have to be sold to farmers who are used to keeping their tractors for decades and repairing the vehicle hardware themselves.”

Here, a tractor’s slow speed is an advantage technologically, Karazuba said, because it means the vehicle can be less “processor-intensive” than a car traveling highway speeds. Furthermore, there is less data to train tractor AI compared to the massive data sets available for cars.

For autonomous ships and other watercraft, the complexity increases due to the difficulty of enabling the cameras and other sensors to function despite interacting with water, salt, humidity, adverse weather, or even just a moonless night. “When you have a car, you can illuminate the area in front of the car with headlights,” Karazuba said. “On a ship, you can point a light forward but you can’t bathe the entire area like you can with a car. That removes hours a day of visibility depending on where you are on the globe. The same is true for aircraft.”

While the industry is heavily invested in autonomous automotives from both a life-saving and profit-generating standpoint, the ecosystem of semi-autonomous vehicles goes far beyond self-driving cars.

“It’s very much like the race to the moon,” said Siemens’ Fritz. “We created so many technologies that at the time were specifically to get our people to the moon, but since then have branched out to change our lives in so many different ways.”

Fritz pointed to a wireless security camera that is powered by a solar panel and is connected to a cellular system, a product similar to the machines being implemented or considered in various smart city projects. The camera has the ability to differentiate between a human, animals, and other moving objects like a leaf, which speaks to the progress made in AI object detection and classification. He also noted that OLED films can be placed inside the windows of a vehicle to create the feeling of being immersed in an environment far different than the real one outside. “None of us can predict the future, other than the fact that it’s going to be very different, because of all these technologies that are being pioneered now for autonomous vehicles,” he said.

Leave a Reply

(Note: This name will be displayed publicly)