by Confidence and safety – of consumers, the public and governments – will be crucial issues for AI and autonomous technology in the coming year. However, earning that trust requires fundamental innovation in the way autonomous systems are tested and evaluated, according to Shawn Kimmel, executive director EY-Parthenon Quantitative Strategies and Solutions at Ernst & Young LLP. Fortunately, the industry now has access to innovative techniques and new methods that promise to transform the field.
The new autonomy environment
Automation has historically been touted as a replacement for “boring, dirty, and dangerous” jobs, and that continues to be the case, whether working in underground mines, maintaining offshore infrastructure, or, triggered by the pandemic, in medical facilities . Keeping people out of harm’s way in sectors as important and diverse as energy, natural resources and healthcare remains a worthwhile goal.
But self-directed technologies are now going beyond those applications and finding ways to improve efficiency and comfort in everyday spaces and environments, says Kimmel, thanks to innovations in computer vision, artificial intelligence, robotics, materials and data. Warehouse robotics has evolved from glorified trams that move materials from A to B to intelligent systems that can move freely in space, recognize obstacles, change routes based on inventory levels, and handle delicate items. In surgical clinics, robots excel at microsurgical interventions, where the slightest human tremor has a negative impact. Startups in the autonomous vehicle sector are developing applications and services in niches such as mapping, data management and sensors. Robo-taxis are already in commercial use in San Francisco and are expanding from Los Angeles to Chongqing.
As autonomous technology enters more and more contexts, from public roads to medical clinics, safety and reliability simultaneously become more important to prove and more difficult to ensure. Self-driving vehicles and unmanned aerial systems have been involved in accidents and casualties. “Mixed” environments with both human and autonomous agents have been identified as novel security challenges.
The expansion of autonomous technology into new areas brings with it a growing number of stakeholders, from device manufacturers to software startups. This “system of systems” environment complicates testing, security and validation standards. Longer supply chains, combined with increased data and connectivity, create or exacerbate security and cyber risks.
As the behavior of autonomous systems becomes more complex and the number of participants increases, security models with a common framework and terminology as well as interoperable tests are required. “Traditional systems engineering techniques have reached their limits in autonomous systems,” says Kimmel. “A much larger set of requirements needs to be tested as autonomous systems perform more complex tasks and safety-critical functions.” This need, in turn, fuels interest in finding efficiencies to avoid increasing test costs.
That requires innovations like predictive safety performance measurements and preparing for unexpected “black swan” events, Kimmel argues, rather than relying on traditional metrics like mean time between failures. It also requires ways to identify the most valuable and impactful test cases. Industry needs to refine its testing techniques without making the process overly complex, costly, or inefficient. To achieve this goal, it may need to manage the set of unknowns in autonomous systems’ operational mandate and reduce the test and safety “state space” from a semi-infinite to a testable set of conditions.
test, test
The toolkit for safety, testing and securing autonomous systems is constantly evolving. Digital twins have become a development factor in the field of autonomous vehicles. Virtual and hybrid “in-the-loop” test environments enable system-of-system testing that includes components developed by multiple organizations throughout the supply chain, reducing the cost and complexity of real-world testing with digital ones Extension.
