Source: ECN article
ECN invited leading experts from various automotive component suppliers to share their views on autonomous vehicle challenges.
Q: What is the largest challenge that autonomous vehicles will have to address before becoming ubiquitous?
By: Ron Demcko, Fellow, AVX
Each year, approximately 40,000 people die from auto-related accidents in the U.S. A large percentage of these accidents are due to distracted drivers and driver error. Autonomous vehicles have great potential to dramatically reduce these numbers, not to mention save the economy from cost of injuries and lost work.
Despite the great promise of such advantages, several factors must be addressed before autonomous vehicles are commonplace. Among them are:
- Winning public and individual trust. All too often, people make comments like, “I would never trust technology enough to let it take over driving.” The fact is that sensors and properly configured machines can make faster, consistent, and emotion-free decisions than humans. In reality, not all accidents can be eliminated—even with autonomous cars.
- Continuous technology and standards evolution. Today’s technology will evolve into even higher performing systems. Great benefits will occur from standards dictating acceptable levels of safety performance. Well-drafted standards could very well increase the rate of development, and reduce overall system cost per vehicle.
- Streamlined government regulations. Certification performance methods and levels need to be published to identify what is expected in terms of performance requirements and how systems will be tested to reflect real-world situations.
- Hardware reliability and service certification. The same people who would never trust technology to take over driving also question system reliability. Autonomous vehicle certification methods must be established for vehicles that are multiple years old. This would address complaints that these vehicles are only going to work for a few years before needing an upgrade.
Autonomous vehicles have the potential to change everything from driver, passenger, and pedestrian safety, to the delivery and distribution of goods. With this much at stake, government and industry sources are working to address any and all issues currently delaying the wide-scale deployment of autonomous vehicle technology.
Co-chair of the Symbiotic Autonomous Systems Initiative, IEEE-FDC, & Head of the Industrial Doctoral School, EIT Digital
Allowing an autonomous vehicle to replace a human driver at the wheel is a matter of trust, which (in the end) is a matter of perception. Because we are not used to self-driving cars, we don’t trust them, and whatever happens that reinforces doubt is accepted as proof of our distrust. For example, there have been a number of headlines where self-driving cars have been involved in accidents. Yet, in most of those instances, it was found that fault was not due to the autonomous driving system. Nevertheless, those types of news stories incite fear and reinforce peoples’ doubts.
Consider that on any given day, there are thousands of automobile accidents, and the possibility a majority of them could have been avoided if the cars were self-driving. Unfortunately, given the low number of self-driving cars on the road, statistics are scarce, and there’s been no real study on the possible avoidance of autonomous and human factor accidents.
In reality we do not have a clear definition of accountability in the case of accidents. For example, accident fault and accountability could be placed on the car manufacturer, self-driving subsystem, or infrastructure that was feeding data to the self-driving car.
Insurance companies are still very wary, and there is a lack of clear regulations.
Over the next decade, self-driving cars will become a lot safer than normal cars, to the point that a few countries will encourage and incentivize citizens to adopt these vehicles. I see this need for a top down push for autonomous vehicles to win the mass market.
Pricing is a factor, but broad adoption will drive prices down, new technologies beyond expensive lidar will become available, shared intelligence will increase safety and decrease cost. Still, the real stumbling block, as I see it, is overcoming the trust aspect.
Senior Technical Marketing Manager, Macronix America, Inc.
Engineers developing systems for autonomous vehicles face myriad challenges to ensure security in the latest vehicles. Securing critical information required for their safe operation is a key focus area for designers.
Automobiles’ electronics systems are quickly evolving to a centralized compute architecture—an evolution not unlike that of smartphone platforms. This new centralized architecture also implements a highly complex, parallel-processing model typical of today’s artificial intelligence applications. This helps facilitate vehicle-safety technology, like advanced driver-assistance systems.
As with any emerging architecture, especially where people’s lives are at stake, ensuring security presents an important challenge. Macronix and other leaders in the autonomous-vehicle ecosystem are developing technology, notably in non-volatile memory, designed to tackle these security challenges.
Because today’s on-chip embedded flash densities are too limited for automotive systems, engineers favor a combination of on-chip and discrete memory. The link between a system’s host and storage needs to be secure to thwart “man in the middle” attacks. That storage must also be resistant to physical probing, handle extreme temperature fluctuations and last the lifetime of a vehicle.
Now that designers have at their disposal non-volatile-memory solutions designed specifically for automotive applications, such as Macronix’s AEC-Q100 qualified NOR and NAND flash devices, the next challenge is to determine how they’re utilized. Common uses include system-configuration storage and securely storing software, either in the factory or through remote over-the-air updates. Another is storing cryptographic keys to secure vital communication links, support authentication (ensuring only authorized people gain access to select parts of vehicles’ electronics), and enable permission-based systems. And in case of an accident, non-volatile memory can securely log events leading up to and during the event, facilitating failure analysis.
The challenges continue to grow, but non-volatile memory continues to evolve to meet those challenges.
Director of Advanced Technology Market Development, Molex
Automakers continue to face new challenges as they strive to build next-generation, intelligent vehicles and, as a result, they’re relentless in their demand for new solutions and technologies while they work toward a truly autonomous vehicle. Right now, we’re seeing innovations occur all across the automotive technology landscape—from the rise of new cellular technologies and developments in artificial intelligence to the deployment of smart mobility solutions and the increasing integration of consumer devices.
But there’s one thing they all need to achieve in order to be successful, no matter their specific challenge or niche.
Sensor advancements that support 360-degree imaging around the vehicle are changing daily and with that, comes the challenge of moving substantial amounts of data—both compressed and not. That then begs the question of what network architecture design is most beneficial—keeping architectures distributed or centralizing them? There’s no clear answer and both have tremendous advantages, creating an elastic struggle that, at times, presents a bit of push and pull.
Time synchronization among all sensors with low system latency is a challenge that will inevitably redefine the entire vehicle wiring approach. New networking solutions will likely evolve that can aggregate and consolidate traffic flow for mission critical data.
Agility is another key challenge the industry is working to solve. Single point-to-point solutions are closed and costly. They are inflexible in many ways with new silicon and processing power becoming available regularly. New silicon or added functionality often times require a complete redesign in order to accommodate thermal challenges or new input cable and connector solutions—and those are just a couple of challenges that must be considered and drive change.
New software solutions are yet another subject area the industry is working to address as it relates to flexibility. Software solutions change daily and when you consider the number of systems already deployed, how practical is it to recall vehicles for improvements and potentially safety-related updates? A need for standardized SOTA/FOTA processes with security at its core is critical for widespread deployment and flexibility.
The need for flexibility—in these areas as well as others—is key to the successful development of an autonomous vehicle and to its ubiquitous existence. It’s flexibility in all aspects of system architecture that is the clear enabler for true autonomy.
Automotive Director, Taoglas
Autonomous driving is one of the four major future enabling automotive market megatrends. By 2030, up to 15 to 20 percent of new cars sold could be fully autonomous. The introduction of advanced driver assistance systems (ADAS) will influence and help determine vehicle regulations, systems standards and testing criteria, as well as infrastructure requirements, affecting consumer acceptance and mass market introductions of fully autonomous vehicles. City and urban autonomous taxi services are predicted to be available in the near term. We will see changes in car ownership patterns, with specific car sharing services supporting consumer solution specific needs for commuting, shopping, vacation, and business needs coming next. The trucking industry will also benefit from autonomous driving. No longer limited by driver availability and driver regulations, shipping and logistics operations will be able to optimize fuel and electricity consumption, reduced driver headcount, and 24/7 operations.
The recent technology breakthroughs have been tremendous in sensor and antenna technology, driving industry excitement. Reaching economies of scale of leading edge technologies of dual-band receivers and antennas that deliver both the throughput and precision needed for autonomous driving, will also be needed to achieve dramatically pushing the market acceptance forward.
Fully autonomous consumer vehicles will not happen en-masse for several years. The technologies will be in place much sooner, but the full alignment of regulations, standards, and consumer acceptance, needs to mature as well. By the middle of the next decade, with a new generation of consumers requiring various modes of transportation, we will begin to see the ramp up of Autonomous vehicles and growing consumer adoption and acceptance.
Head of Innovation & Incubation, Elektrobit
It would be easier to address if there was only one challenge, but if I had to sum it up, I would say that the challenge lies in the ability to create a safe, self-driving vehicle that is affordable and can be used in a fully-autonomous mode by anyone, anywhere.
We’re at an interesting point in the evolution of the autonomous vehicle—I believe we have the necessary technologies, or at least understand what we’ll use, but it will take time to put in the work to develop a vehicle that can get passengers anywhere they need to go—safely. The work is testing and simulating driving scenarios, training vehicles using machine learning and analytics, implementing security and mitigation, building the HD maps, etc. Additionally we need at least one more generation of some perception sensors to get into a price range making a private vehicle affordable. There is a lot of work that goes into creating a capable and reliable vehicle, and OEMs, Tier 1s, hardware, and software companies are all working toward that goal over the next decade.
It won’t take a decade for many of us to begin using autonomous vehicles as part of mobility service solutions. In fact, by the time autonomous vehicles are truly ubiquitous, we all will have used them as they’ll first be used in a shuttle, tram, or ride hailing service in some limited geographic situation.
Senior Director of Automotive Business Unit, Xilinx
A number of challenges are preventing autonomous vehicles from becoming ubiquitous.
The main one is the lack of commonality or uniformity of technical approaches that drives commoditization.
Engineers are working to solve the root of the challenge—which is getting a machine to behave as a human. The below chart highlights the three main dimensions to modelling the human ability:
|Senses; sight, hearing, touch||Cameras, Ultrasonic, Radar, Lidar|
In each of these areas, there are a multitude of ways to solve the problems. As a result, the best approach has yet to be achieved, and the technology is still evolving. Today, automakers are still unable to standardize on the sensors set needed, meaning how many cameras, radars, and lidar modules are required to generate enough data to build contextual awareness of the vehicle to its environment.
Let’s examine one particular area: lidar. There are a handful of major automotive Tier 1s that are offering lidar technology to OEMs through a mechanical approach. On another spectrum, if you do a keyword search for lidar on Pitchbook, a database for companies both public and privately funded, you will find that there are 67 start-ups globally that are working on advancement of lidar sensing to increase the technical parameters such as field-of-view, range, and resolution of point cloud.
One reason FPGA technology is widely adopted in lidar is that the fragmentation of approaches makes it difficult for other silicon vendors to achieve high-volume production scale and therefore they are unable to offer ASICs that are optimized for the task at hand.
The lidar industry is seeking a standardization like the old VHS or Beta. Once this is achieved, then, we will be one step closer to making autonomous vehicles a reality.