As automakers rev up their engines in the race for self-driving cars, all routes to the finish line run through artificial intelligence.
But just as automakers have come to accept that succeeding in the 21st century requires adopting new paradigms, AI itself must undergo major transformations in order to fulfill its promise of making our cars smarter, our roads safer, and our future autonomous.
Achieving autonomy will require substantial improvements in how AI technologies handle conditions on the road – but bringing self-driving cars to fruition requires far more than better AI capabilities. Automakers’ bottom lines and the industry’s long-term success will require new efficiencies and adjustments, from product integration to software to energy efficiency.
Far from a pipe dream, these improvements are already underway, with their greatest impact yet to be felt. Here’s a look at four groundbreaking automotive AI trends that will help propel automakers toward autonomy in 2019 and beyond.
#1: Going Beyond Lidar
Lidar, long the undisputed technology of choice for autonomous driving, no longer reigns supreme. While it was initially believed that Lidar could deliver on the promise of AI, sheer experience has demonstrated that Lidar alone is simply not accurate enough.
To be sure, Lidar is indisputably powerful, but it also has some fundamental deficiencies. For example, it often indicates nonexistent “ghost” objects, has a limited range, and cannot perform in all weather. To drive autonomous innovation, Lidar-plus is essential.
Fortunately, the fusion of Lidar and camera data can not only solve Lidar’s biggest shortcomings, but can also offer features beyond the scope of Lidar, such as traffic light detection and road sign recognition.
#2: The Edge is the Future
Automakers are struggling with the consequences of adding AI to their products, mainly due to the technology’s expensive, bulky and power-hungry hardware, which make it quite impractical for many use cases.
As a result, a shift is underway from centralized computer vision systems towards edge computing, which runs on low-power processors next to the sensor, reducing wiring and communications costs. This pioneering technology can be mainstreamed in an affordable way thanks to highly-efficient algorithms that can run on slim processors without compromising on quality or performance.
#3: Goodbye to Black Box Hardware
Hardware-based AI takes years to develop, and when it’s finally ready, proves quite inflexible whenever important fixes or new features are necessary. Updates that require fundamental modifications sometimes can’t be done until the next generation of hardware is available. In the automotive industry, that can take years.
Additionally, leading automakers and component suppliers who prefer customized AI aren’t able to add their secret sauce to the rigid hardware and must settle for one-size-fits-all solutions. In response, the industry is now shifting from hardwired technologies to more flexible, hardware-agnostic, software-based solutions that are quicker to market, support frequent FOTA updates, and can be customized to specific requirements, allowing better fit and differentiation.
#4: Battery-Friendly is a Must
As cars are packed with sensors, minicomputers, and miles of wiring, the power supply allocated for each module is strictly rationed. Automakers using high-end processors for AI computing needs also rapidly discover that cooling these units is highly problematic; active cooling systems and large heat sinks are simply not cost-effective for vehicle edge technology. With electric cars growing more common and battery range becoming a key factor, automotive AI must be battery-friendly to meet OEM requirements. As a result, the industry will increasingly prefer AI solutions with efficient power consumption and heat dissipation.
While OEMs and data scientists have been honing automotive AI for years, 2019 represents a critical juncture on the road to full autonomy. With automakers moving to enhance the technologies and performance of today’s connected cars and ensure a smooth transition to tomorrow’s autonomous vehicles, we may well look back on 2019 as a bridge between the two.