By Michael Belfiore, Contributor
The Indianapolis 500 brings together 235,000 fans for a once-a-year event pitting 33 drivers against each other in a 200-lap, 500-mile test of skill and endurance. But in the fall of 2021, the teams racing on the oval will pit software against software whose engineers will sweat out the race, not in the cockpits of their cars but in their on-site garages and pit stops.
The first-of-its-kind Indy Autonomous Challenge (IAC) will use the same Dallara IL-15 race cars as the famed Indy 500 race and take place on the same speedway. The speed, the skill, the danger, and the prize money—a total of $1.45 million—will all be there, but this time the cars will be on their own.
“If we want to rely on automated vehicles for our well-being, we need to have confidence that they can address situations that emerge that are infrequent but extreme.”
—Matt Peak, managing director of Energy Systems Network and head of operations for the IAC
“If we want to rely on automated vehicles for our well-being, we need to have confidence that they can address situations that emerge that are infrequent but extreme,” says Matt Peak, managing director at Energy Systems Network (ESN) and head of operations for the IAC. “And some of them can be modeled and addressed very well with high speed.” That includes such situations as sudden slowdowns on the highway, cars cutting each other off, and rapidly merging.
Just as conventional auto racing has led to disc brakes, all-wheel drive, and even rear-view mirrors, IAC participants hope to significantly move the needle on self-driving car technology. “Collaboration will happen as an outcome of this work, when we look back and pinpoint this event as having contributed in a meaningful way toward the safety of self-driving cars,” says Madhur Behl, head of IAC team Cavalier Autonomous Racing, based at the University of Virginia.
Since the DARPA Urban Challenge race for self-driving cars concluded with three prize-winning finishers in 2007, major automakers, artificial intelligence (AI) experts, engineers, and investors have banked on the promise of cars driving themselves to free up commuters and other humans to do things like text, read, or just take a nap.
About 40,000 Americans die in car accidents every year. The National Safety Council estimates that distraction causes 8 percent of those accidents and drowsiness another 2 percent. Autonomous cars could be a possible solution, and DARPA and the winners of the Urban Challenge demonstrated that vehicles could drive themselves through city streets without human intervention. But the most ambitious of DARPA’s self-driving car races were in controlled conditions, with a speed limit of 30 miles per hour.
Despite the promise, autonomy for passenger vehicles has remained perpetually five years away. Tesla, with arguably the most advanced self-driving technology on the market, has included the hardware needed for full autonomy in its cars since 2016. But four years on, the actual capability isn’t here yet.
To get to the next level, says Peak, engineers have to test cars going faster, and they have to react to each other in ways they haven’t done before. “If teams traveling 200 miles per hour can navigate around each other and jockey for position without losing control, then perhaps that software and technology are relevant to helping you avoid that piece of lumber that’s fallen off that pickup truck at 65 miles per hour in front of you.”
It’s that kind of edge case, as unexpected events are known in AI circles, that’s so hard to test in the real world. “The head-to-head aspect is really the secret sauce of this,” says Dan Reilly, a member of MIT Driverless, the IAC team based at the Massachusetts Institute of Technology.
A Learning Experience
Universities will field all of the teams in the IAC. That’s by design, Peak says, to keep the competition manageable. They span the globe, from the U.S. to South Korea, Europe, and India, among other regions. For ESN, the nonprofit organizing the race, the competition fits its mission to advance technological innovation and influence policy, according to Peak.
Dallara, the Italian builders of the IL-15, is supplying the vehicles, while Ansys is providing simulation software and putting up some of the prize money. Other sponsors will provide sensors and other components to enable full autonomy. The competition will come down to the software—to be written by the teams—to make all the hardware work together.
MIT Driverless, for example, plans to compete on the way it brings sensor data together to enable their car to react as quickly as possible to ever-changing race conditions. “When your sensors only give you images 30 times a second and you’re going 100 miles per hour, it’s many, many yards of distance you’re covering while being completely blind,” says team member Nick Stathas. The team’s software will have to compensate with as much data from as many sensors as possible delivered in real time to the onboard processors.
“There are perception challenges in racing,” Behl agrees. “It’s different to get a camera feed and do object detection when you are at 14 miles per hour compared to 200 miles per hour. Your camera images get blurred. You don’t get sharp pictures. Your LIDAR sensor is vibrating all the time, so you’re not getting clear LIDAR data.” Again, the team’s software will have to compensate.
Both the MIT and UVA teams have about 30 members, a mix of undergraduates, graduate students, and faculty advisors. Each has participated in sub-scale electric vehicle competitions at low speed.
Still, none of the teams has ever raced at the 150-200 miles per hour speeds of an Indy 500, and none has raced other vehicles while trying to avoid skidding off the track or colliding with each other. For that reason, the IAC promises to be an event like no other, with the potential to move the needle on autonomous driving like never before. Ladies and gentlemen, start your processors.