Multiple tech issues stall AV deployment - Embedded.com

Multiple tech issues stall AV deployment

An autonomous vehicle ecosystem is emerging, but AV companies remain far from developing software operators anywhere near human drivers.

Despite progress, autonomous vehicle technology is widely perceived as years away from commercialization. Contributing to that perception is the AV industry’s overpromising and underperforming over the last five years. Many technology issues must still be resolved before deployment takes off.

We review progress and offer perspectives on the current status of AV development while considering what is needed for future deployment. We’ll also cover many of the emerging road AV use cases.

AV technology safety standards have been introduced. The emerging specs are critical to deployment timelines and must be integrated with next-generation AV systems.

Road-use legislation for AVs is the next step, and will be required for successful AV deployments. Compliance is expected to pose significant challenges as new safety requirements are added.

The table below summarizes the current state of the AV industry, technologies and use cases in the most obvious categories. It addresses three timeframes, divided into four segments.

click for full size image

Historical perspectives

AV development and testing stages for leading companies extended from 2015 to 2020. Some started earlier and others are still testing. AV use cases are discussed first since they greatly influence other considerations.

AV Use Cases

Initial research focused on the most promising applications. Initially, ride-hailing took off in China, the U.S. and other regions, making robotaxis the most desirable early AV deployment scheme. In Europe, fixed-route AVs attracted the most attention since ride-hailing growth was restricted in most cities. U.S. and China were the key markets for autonomous truck development and testing.

Understanding technological complexity, customer needs and other use-case factors required much research. Generating training data for testing AV software was critical. Virtual AV testing via simulations has increased in importance, and remains crucial for safety.

Road Use Legislation .

There has so far been little government oversight. Test permits are needed in most cases and safety drivers are required for all AVs transporting passengers. The goods-only AVs also needed road permits, along with autonomous trucks that began testing during this period.

Some U.S. states issued test permits, California’s being the largest program. China started robotaxi testing in the middle of this timeframe. Europe was primarily testing fixed-route AVs and some sidewalk goods AVs.

By 2019, more than 65 companies had received California DMV permits to test AVs. Permits for a total of 567 test vehicles were obtained, driving nearly 2.9 million miles that year. In 2020, the number of permits reached 668 test vehicles, but miles driven dropped to 1.99 million owing to the pandemic. AVs needed a safety driver for testing in California during this timeframe.

Safety Standards.

Tech safety focused primarily on functional safety based on ISO 26262 and its four ASIL (Automotive Safety Integrity Level) stages. Work also started on technical safety standards for AVs introduced later.

SAE defined six levels of driving, with L3, L4, L5 representing autonomous driving. Those levels were widely used. SAE also introduced multiple important concepts:

  • ODD, for Operational Design Domain, defines the environment where an AV can operate.
  • OEDR, or Object and Event Detection and Response, monitors and responds to the driving environment.
  • DDT, or Dynamic Driving Task, covers AV driving, performing OEDR and avoiding crashes.
  • ADS, for Automated Driving System, refers to the computer system that performs the DDT.

AV Technology

Technology development received the bulk of attention during this period with multiple new technologies emerging. The software driver is the most complex component since it needs to accomplish all DDT functions. Sensors and vision software are used to accomplish OEDR functions.

Computer hardware to run all the software required much higher performance than available in this timeframe, a challenge tackled by Nvidia and other processor designers.

AI technology advanced rapidly during this stretch, perhaps so fast that current AI improvements were overestimated.

Current status

AVs are currently in the deployment trial stage for most AV use cases. The trial stage for some started in 2018. This phase will probably last through 2025 or longer for some use cases in some regions.

Use Cases.

Use cases have evolved, and their importance has change to some degree. The pandemic increased the importance of goods AVs with autonomous trucks becoming much more desirable due to truck driver shortages and big increases in online orders and deliveries.

Robotaxis remain important with driverless trials starting in U.S. and China and more trials expected over the next few years. California has granted permission for seven companies to conduct driverless testing. Three have received authorization for deployment of AVs in specific areas—Cruise, Nuro and Waymo. Nuro is testing goods-only AVs while Cruise and Waymo focus on r obotaxis.

Autonomous trucks are generating revenue by delivering goods while in testing mode with safety drivers. This use case primarily uses a hub-to-hub deliveries scheme where most of the driving is on highways with short trips from a highway to a warehouse for each roundtrip.

Sidewalk AVs are delivering food on many campuses and groceries in some cities in U.S., Europe and China.

Road-Use Legislation .

AV legislation remains limited, but much planning is underway. France and Germany have taken the first steps toward crafting AV legislation. Others are in the process of passing new legislation, including China, the U.K. and in the U.S.

More test permits are available in more countries. Europe is finally becoming more active in robotaxi testing, with trials in Munich, Paris and other cities. A new ISO standard, discussed below, will increase fixed-road AV testing and deployment with Europe likely leading the way.

Safety Standards.

Safety standards have advanced in the last year or so. The ISO 22737 LSAD is especially important for some AV use cases. LSAD, or Low-Speed Autonomous Driving, will be used for many fixed-route AVs, good deliveries and similar scenarios.

ISO 21448, or Safety of the Intended Functionality, extends functional safety to AVs. The spec covers functional safety vulnerabilities occurring without system failure. (ISO 26262 covers system failures.)

UL 4600 addresses L4-L5 AVs, including all aspects of autonomous driving development. UL 4600 was developed by Underwriters Laboratories and Edge Case Research. It extends ISO 26262 and 21448 to AVs based on their current state and sensing their operating environment with no human intervention.

IEEE P2851 is leveraging Mobileye’s AV driving algorithms for avoiding crashes—the Responsibility-Sensitive Safety (RSS) model. RSS is a formal, mathematical model that applies a technology-neutral approach for avoiding crashes. The standard is called Assumptions for Models in Safety-Related Automated Vehicle Behavior.

The SAE J3016 terminology standard that defines AV levels and concepts is attracting growing criticism, perceived as a safety standard when it only defines AV terminology. Our colleague Colin Barnden provides an excellent overview on this and other AV safety topics.

It is likely that SAE will update its six levels from ADAS to AVs. A proposal by Phil Koopman of Carnegie Mellon University is gaining considerable support based on Koopman’s four simplified automation modes:

  • Driver Assistance: Supports human driver.
  • Supervised Automation: Human driver monitors and fills in capability gaps.
  • Autonomous Operation: No human driver responsibility for safety.
  • Vehicle Testing: Driver mitigates risk due to potential design defects.

AV Technology

AV technology continues to advance—some facets more than others. Software drivers are improving, some covering multiple AV use cases. Edge cases or new driving situations not previously seen remain a tough problem, requiring considerable advances for acceptable safety.

Sensors are advancing with improved technologies becoming available. A lidar technology battle between time-of-flight (ToF) and Frequency Modulated Continuous Wave (FMCW) technologies is underway. ToF is the established technology used by most AV startups. FMCW has key advantages such as minimal interference, excellent speed measurement and longer range. Cost, however, remains an issue. FMCW looks like the long-term winner.

A new generation of radar technology is emerging that will greatly improve sensing capabilities via 4D or imaging radar. 4D radar brings overlapping functionality with lidar and is the long-term radar leader.

Elsewhere, far infrared (FIR) is another promising AV sensor technology. FIR can detect warm bodies better than other sensors and will be crucial for detecting pedestrian humans and animals.

AI technologies continue to improve, but the rate of advances seems to have slowed—at least for AVs. One issue is the so-called black box problem: It remains difficult to determine how AI models make decisions for AV driving. Such obscurity remains a big problem for AV usage. Better solutions are needed.

Future perspectives

The safe deployment stage will vary by AV use case, with a timeframe from 2024 to 2030 and possibly later—based on the plans of leading players. The next section reviews the chart above.

AV Use Cases.

Use cases with relatively simple traffic patterns including low-speed trips will see safe deployment first. This includes AVs for fixed and variable routes in many mass-transit deployments. Goods-only delivery AVs will also see early deployment with sidewalk AVs as the simplest version operating at walking speed.

Road based goods-only AVs such as Nuro, should also see early deployment. Nuro’s third-generation AVs include an external airbag to improve safety for pedestrians.

Robotaxis are on the way to deployment, but only in restricted areas of select cities. Expansion will depend on road-use legislation. Future operational safety for robotaxis will also have a big impact on future expansion. A few deadly crashes could quickly snowball into major problems for all robotaxi operators—even if only a few have crashes result in injuries or deaths.

Consumer AVs will leverage robotaxi development and deployment, but by a lag time of at least five years. Consumer AVs are likely to have limited ODDs in a pattern similar to robotaxis. The biggest factor in determining consumer AV deployment will be future road-use legislation.

Some autonomous truck applications will be deployed by 2025, with hub-to-hub leading the way. Autonomous trucks promise to improve operation even with a safety driver, especially given current driver shortages and better utilization of a driver hours between legally required rest times. Driver efficiency can shorten trip times for goods such as fresh produce with significant cost advantages.

Road Legislation .

Significant road-use legislation is likely in many countries by 2025. These laws will have profound impact on all AV use cases, perhaps delaying AV deployment.

Several tough decisions must be made as part of these deliberations since new AV legislation will be based on minimal data for crafting specific provisions. Among them is specifying the starting point of AV safety compared to human-operated vehicles, and how to measure the evolution of that relationship. AVs are improved via software updates that improve safety. Gauging safety improvements against a particular set of human drivers remains an open question

The U.K. effort to pass AV legislation contributes much to the debate. For example, U.K. researchers concluded AV safety is a political decision best left to Parliament.

Little discussed so far is the tremendous revenue potential of future AV markets for companies and governments. That revenue potential will undoubtedly be factored in AV road-use legislation. Early AV deployment will also confer technology and usage leadership that could be leveraged in other countries as AV deployment advances.

Safety Standards.

AV tech standards are in good shape: ISO 21448 and UL 4600 have been approved, and IEEE P2851 is on the way. These standards must be built into next-generation AV systems, adding considerable safety.

New technology standards or at least updates of existing standards are likely. One question is whether there will be specific tech standards for vehicle categories such as autonomous trucks, goods-only AVs or others?

Future legislations may also require new or updated standards in the 2025 timeframe. This may have the biggest impact on future AV tech standards.

AV Technology.

Continuous AV technology improvement is required. The software driver will rely on AI technology advances and improvements from expanded machine learning and data from testing and operations.

Sensor hardware and software technology such as FMCW lidar, 4D radar and FIR are entering volume production. All sensors will see cost declines and improved reliability, with lidar becoming more affordable.

Computing performance will also advance as cost declines. Chips with AI accelerators will help increase importance. Some AV companies will develop their own chips.

Bottom line

AVs must overcome multiple challenges before volume deployment occurs. Technological complexity remains daunting: developing and proving that software drivers can outperform human drivers remains a major hurdle.

AV use cases are becoming clearer with some easier to master than others. Deployments with uncomplicated traffic patterns and low speeds will see early volume deployment.

AV technology standards are progressing. It is now up to AV companies to quickly integrate these standards into their platforms.

Perhaps the toughest and probably most controversial decisions will be made by politicians forging AV road-use legislation. They must weigh AVs safety against the ability of human drivers. They must also specify implementation of new laws, how AVs will be authorized for public use and how AVs remain authorized during their driving lifetimes.

>> This article was originally published on our sister site, EE Times.


Egil Juliussen has over 35 years’ experience in the high-tech and automotive industries. Most recently he was director of research at the automotive technology group of IHS Markit. His latest research was focused on autonomous vehicles and mobility-as-a-service. He was co-founder of Telematics Research Group, which was acquired by iSuppli (IHS acquired iSuppli in 2010); before that he co-founded Future Computing and Computer Industry Almanac. Previously, Dr. Juliussen was with Texas Instruments where he was a strategic and product planner for microprocessors and PCs. He is the author of over 700 papers, reports and conference presentations. He received B.S., M.S., and Ph.D. degrees in electrical engineering from Purdue University, and is a member of SAE and IEEE.

Related Contents:

For more Embedded, subscribe to Embedded’s weekly email newsletter.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.