Despite a pandemic that continues to punish global economic markets, not to mention vehicle manufacturers’ and suppliers’ product plans, many heavy-duty truck makers are not relaxing on efforts to bring high-level (SAE Level 4 or 5) autonomous trucks to market. Partnerships with self-driving tech startups can be a big part of the roadmap, the latest example coming in mid-July as Navistar announced it will co-develop SAE L4 trucks, targeted for production by 2024, with San Diego-based TuSimple.
The producer of International commercial trucks and IC Bus school and commercial buses also has taken a minority stake in TuSimple. These moves follow a two-year “technical relationship” with the startup, which operates a U.S. fleet of 40 self-driving trucks that move freight between Arizona and Texas for companies like UPS and McLane Company. TuSimple plans to demonstrate completely driverless operations in 2021.
“TuSimple and Navistar began joint development of pre-production units in 2018, and now we are kicking-off a full go-to-market production program,” said Cheng Lu, president of TuSimple. “With [our] combined expertise, we have a clear path to commercialize self-driving Class-8 trucks at scale.” But Starsky Robotics, which shut down operations in March 2020 after four years of developing autonomous-truck technology, is a stark reminder that the path to fully autonomous commercial vehicles is not an easy one.
Starsky co-founder and CEO Stefan Seltz-Axmacher attempted to explain in a blog post how and why his company went from operating the first fully unmanned truck on a “live” highway in 2019 to being defunct within a year of doing so. Beyond lamenting that trucking companies are not great technology customers and investors and others prefer “cool” features over mundane safety – “We couldn’t figure out how to make safety engineering sexy” – he said he believes artificial intelligence (AI) not being truly ready for primetime is the main problem facing the AV industry.
“Supervised machine learning (ML) doesn’t live up to the hype. It isn’t actual artificial intelligence akin to C-3PO, it’s a sophisticated pattern-matching tool,” Seltz-Axmacher wrote. “It’s widely understood that the hardest part of building AI is how it deals with situations that happen uncommonly, i.e. edge cases. In fact, the better your model, the harder it is to find robust data sets of novel edge cases.
“Additionally, the better your model, the more accurate the data you need to improve it. Rather than seeing exponential improvements in the quality of AI performance (a la Moore’s Law), we’re instead seeing exponential increases in the cost to improve AI systems – supervised ML seems to follow an S-Curve.” Despite such challenges, many automated-driving technology experts continue to project commercial vehicles as the most promising early-adoption sector for high-level automation.
Redundant system setup
“The autonomous self-driven commercial vehicle will come – for sure,” asserted ZF’s Dirk Wohltmann, engineering director, Americas, providing a Tier-1 supplier perspective in a recent SAE technical webinar on the topic. But when and how it can be realized depends heavily on the usage environment, or Operational Design Domain (ODD), he added.
In controlled environments that can limit or eliminate unexpected obstacles and other unpredictable vehicles/scenarios, such as mining sites, ports and industrial areas, highly automated or even fully autonomous CVs already can be found. But for long-haul trucking, where there’s perhaps the “biggest need and highest efficiency” potential for autonomy, according to Wohltmann, the operating environment is anything but predictable.
Failure detection and the reaction strategy must be robust and invariably reliable to instill confidence that unmanned heavy trucks are safe for public roads. “The system has to reliably detect gaps, sensor or other faults and react accordingly,” he said. “With increasing autonomy, the system needs to take over more and more the diagnostic and especially the reaction.”
A well-defined redundant system architecture is necessary for this strategy, and the systems must be highly connected and centrally controlled. “Each sensor, all communication lines, every controller and the related actuators need to have a backup to take over in case of single individual faults,” Wohltmann said. Using already-available components and actuators can help save weight and complexity compared to incorporating additional actuators. For example, active steering can carry a backup electronic control and actuator, or it could be taken over by the existing service brakes with “steer by brake.”
Incorporating the trailer into the failsafe and redundancy concept is vital to realizing the full potential of an autonomous-vehicle system setup, according to Wohltmann. “If today the driver might only apply the rear axle brake as a last resort in case of service brake failure, the automated system will integrate the front axle and trailer brakes based on ‘knowledge’ of performance and status and loading setup,” he said. “Experience from EBS and brake balancing can be used for a safer and more efficient holistic braking setup and utilized as redundancy.”
Avoiding ‘release paralysis’
“We must acknowledge that the release of a Level-4 truck is a distinguished challenge, one that requires us to approach it with a combination of known methodologies and adopting new ones,” said Suman Narayanan, director of engineering for the Autonomous Technology Group, Daimler Trucks North America. His team is engineering a Freightliner Cascadia L4 platform with fail-operational systems to execute virtual driver commands.
Categories of preventive quality and risk assessment that must be properly addressed include product safety and reliability (ISO 9001 and SAE J1739), functional safety (ISO 26262 and SAE J2980), operational safety (ISO/PAS 21448 and SAE J3018) and cybersecurity (ISO/SAE 21434 and SAE J3061). Safety of the Intended Functionality (SOTIF) is in the final stages before it can be released, and it will play a critical role in development and release of autonomous vehicles, according to Narayanan.
The SOTIF standard (ISO 21448) applies to emergency intervention systems and advanced driver-assistance systems (ADAS) that could have safety hazards without system failure – i.e., scenarios involving inadequate situational awareness or decision-making.
On the testing and validation side, engineers are confronted with a “unique challenge”: how to assess the readiness of automated technology. Narayanan offered one possibility, a methodology long applied by NASA that systematically addresses technology readiness level (TRL), thoroughly assessing a technology’s maturity before it can be safely deployed among the public (see chart in gallery). “The one thing that I find fascinating is the potential ‘release paralysis’ that a chief engineer could face if they do not approach this in the most objective way,” he said.