Autonomy Q&A with dSpace's Moudgal
Automakers and new mobility companies—and their suppliers—are scrambling to develop automated driving systems for Level 3 and beyond. During our Autonomous Vehicle Technology webinar in February, experts from dSpace, Magna, and KPMG looked at the major technology trends and challenges of the fast-growing industry and how engineers are developing, simulating, and testing systems that are soon reaching production. After the webinar, Vivek Moudgal, Vice President of Sales at dSpace Inc., provided further insight and answered a few remaining questions unanswered after the webinar. (Follow this link to view the on-demand version of the webinar, "On the Road to Full Autonomy.")
Moudgal: It should be kept in mind that, for a number of years after fully self-driving vehicles are on the road, they will need to share the road network with human-driven vehicles. Year after year, the accident data that gets published reports that 90% or more of the accidents caused are due to human errors—from DUI (driving under the influence) to distracted driving (texting, not having their eyes on the road) and other such situations. Self-driving vehicles will be programmed to follow local traffic laws and norms, with no speeding; no tailgating; automatically reducing speed in wet, slippery, low visibility conditions; no cutting off other vehicles on the road; and many other human-driver bad behavior.
Once we get to a stage when a majority (if not all) of road vehicles are self-driving, the chances that they get into crashes will be greatly reduced and more of a case of system failure. This is where such issues as redundancy, sensor quality, and fusion algorithms will make the difference.
Given that the quality and number of sensors on the vehicle, as well as the computing and data storage capacity, all contribute to the cost of the system, it is highly possible that the different class of vehicles will have different levels of fault tolerance in their design, leading to different levels of performance under failure conditions. But, in general, the goal will be for the vehicle to consistently execute "safe" maneuvers under all situations.
Also, today’s vehicles incorporate a lot of weight for safety purposes such as crumple zones, door-panel reinforcements, etc. If future self-driving road vehicles are all observing speed limits and following other traffic rules—thus reducing the chances of accidents—it will be possible to take a lot of that “extra” weight out of them. This will enable a switch to smaller engines, motors, and other equipment and increase in fuel economy and reduction in emissions.
What follows is an edited Q&A with Moudgal, of the few remaining questions unanswered just after the webinar.
Question: As you say, we're close to seeing cars driving themselves around Phoenix with no one at all in the car. Is there a way to (re)take control of the car remotely?
Moudgal: In a way, this is already happening (see https://www.azcentral.com/story/money/business/tech/2017/02/21/uber-self-driving-cars-arrive-arizona-tempe-debut-scl/98208998/). There are also self-driving cars on the road in some cities with a “safety driver” not doing the normal driving tasks, but as a backup in case the automated-driving system was to encounter a problem. In other cities, there have been pilot programs that have been implemented to use self-driving vehicles to deliver products to consumers (Dominos and Ford had a program to deliver pizzas to consumers using driverless vehicles). Cybersecurity is a major concern for the makers of self-driving vehicles [and this includes remotely controlled versions]. This by no means is solved and may not be for quite some time. In addition to AI (artificial intelligence) topics being discussed and developed, cybersecurity is the other major area of research keeping people up at night.
Question: How much logging/auditing do you foresee on the car for analysis in a failure?
Moudgal: Today, self-driving vehicles on the road are collecting many dozens of channels of data continuously as they are still in the development phase. When they do go into production, in my opinion it is highly possible that there may be “flight-recorder” functionalities that are built in to record certain data streams—not necessarily to find fault, but to understand the situation that may cause a problem and to use the information to prevent the issue from reoccurring.
Question: What are the largest real-world [AV] obstacles currently? Snow/lane identification was still one of the largest issues. Is that still the case?
Moudgal: With regard to snow and lane-marking issues, the combination of camera, radar, and LiDAR sensors, along with GPS and high-definition maps, makes it possible to very accurately localize a vehicle’s position. Even if lane markings are hidden due to snow or wear, as long as accurate map data are available to the vehicle, it should be able to navigate its way to its destination safely. Of course, there is the issue of getting reliable GPS signal in all areas (urban canyons with their multiple reflection points issues, in tunnels, etc.), but here deployment of INS systems combined with map data helps out. The bigger challenge is with fleet turnover, the need for these self-driving vehicles to coexist with vehicles driven by humans on the same road, and being expected to react to unpredictable human behavior.
Question: China has a very low HR costs; any recent word on their progress in this technology?
Moudgal: The switch to AV is far beyond the low cost of HR. It involves safety, the need to reduce pollution, reducing congestion, etc. A number of Chinese companies, such as Baidu, are investing greatly in this technology.
Question: Would weather and ambient temperature conditions affect performance?
Moudgal: Given that all automotive electronics have to operate consistently in a wide range of environments, the electronics being designed for self-driving vehicles will eventually need to meet the same minimum performance parameters. This said, it is highly advised to slow down in wet, slippery weather. So if the question is WRT vehicle speed performance, then it is highly possible that the vehicles will be programmed to operate in a “safe” manner, slowing down when the computer detects a loss of traction, white out conditions, etc.
Question: How do you expect an AV do operate in the snow and ice or hydroplaning in rainy conditions? A human can correct steering in snow and ice. How will this happen in an autonomous vehicle?
Moudgal: This is where AI will play a role. Today, features—like traction control, electronic stability control, emergency brake assist, and others—greatly aid the driver in safe driving. Transferring this knowledge to the computer will be key. One thing to keep in mind is that once the computer is tuned and tested, it will not tire after multiple hours on the road. Statistics of accidents have shown that accidents are more likely to happen toward the end of long drives when a human driver may be getting tired and thus their reaction may be slower or amplified.
Question: With driverless vehicles, it is claimed there will be a new billion-dollar economy as drivers will be able to work and be productive on their commutes. Can you comment?
Moudgal: This is an interesting question, and there may be many opinions. I currently have a 45-60 min drive to work each way. During that drive, I have witnessed other drivers doing many non-driving activities, including eating, grooming, applying makeup, reading documents, and texting. Even if we do not add a billion dollars' worth of new productivity but can continue to do these activities in a safe manner because the task of driving is being handled by a capable system, we will greatly increase productivity, and possibly also decrease insurance expense by reducing accidents.
Question: Where do you see the accountability of accidents shifting with higher levels of autonomy?
Moudgal: The responsibility to do due diligence in thoroughly testing the self-driving vehicle product will be with the vehicle producer—the OEMs.
Question: Do you think standards can be developed for modeling and simulating inclement weather for the purpose of test and evaluation?
Moudgal: Standards are possible to validate against the real-world expectations. However, simulation of environmental conditions is not new. It has been done for many years when using HIL (hardware-in-the-loop) for testing ABS, ESP, and other ADAS systems. SAE and other professional bodies are continuously writing and improving on standards to account for changes in technologies and coverage. The ISO 26262 standard also is a method that describes how testing should be conducted to ensure that safety-critical systems have a predictable behavior under all conditions.
Question: How will ride-hailing/sharing be adapted for rural areas? Or is autonomy more of an urban or highway technology?
Moudgal: Ideally self-driving vehicles will address all areas—urban, rural, and in-between. However, given that the highest level of congestion, crowding, and lack of parking issues occur in urban areas, this is where the benefit of self-driving technology will be best observed.