The planned Indy Autonomous Challenge—taking place in October 2021 in Indianapolis—is intended for 31 university computer science and engineering teams to push the limits of current self-driving car technology. There will be no human racers sitting inside the cramped cockpits of the Dallara IL-15 race cars. Instead, onboard computer systems will take their place, outfitted with deep-learning software enabling the vehicles to drive themselves. 

In order to win, a team’s autonomous car must be able to complete 20 laps—which equates to a little less than 50 miles in distance—and cross the finish line first in 25 minutes or less. At stake is a $1 million prize, with second- and third-place winners receiving a $250,000 and $50,000 award, respectively.

As Matt Peak, the managing director of the event’s organizer, Energy Systems Network, explained, the competition will help researchers discover how to create autonomous vehicles that can handle so-called edge cases when driving. Edge cases, or atypical events, can cause deep-learning systems to stumble because they haven’t been trained to take them into account. 

For instance, “fast-moving race cars with obstacles coming at them at lightning-quick speeds,” requiring vehicles to “joust and maneuver” through the track, represent a “quintessential edge-case scenario,” Peak said. 

The self-driving car race is part of a larger legacy of the Indianapolis Motor Speedway’s relationship to technological milestones in the auto industry, Peak explained. For decades, researchers have tested cutting-edge auto technologies by racing vehicles at high speeds on the more than 100-year-old circuit. For instance, one race in 1921 tested the capabilities of four-wheel hydraulic brakes in a race, while another race in 1993 tested crash-data recorders; these technologies have since become standard in conventional automobiles. 

Researchers use racing as a testing ground for auto technologies “because it’s so extreme and binary,” Peak said.

“There are fewer shades of gray when it comes to margins of error,” he said. It’s much easier to steer a car to safety if it loses control at speeds of 15 mph, “but it’s catastrophic if you are racing at full speed,” he said.

Currently, a team of Clemson University researchers is designing the specifications for the modified Dallara race cars that all competitors will use, said Robert Prucka, an associate professor at Clemson’s department of automotive engineering. The competitors are responsible for developing the deep-learning systems that will pilot the cars. 

He said the cars will likely have powerful engines falling in the “375- to 400-horsepower range.” For comparison, the average automobile has a 120-horsepower engine. And instead of one car battery, these race cars will likely have several to supercharge their computerized innards, Prucka said.

These race cars will also have advanced camera systems, radar technology, and lidar technology, which uses lasers to scan the surroundings of a car in order to quickly generate 3D maps, so the vehicle can navigate through roads while avoiding obstacles. While startups and auto companies have typically chosen one particular navigation technology like lidar to power their self-driving cars, these researchers can use all of the latest systems. Prucka said the event’s organizers did this because some universities specialize in one navigation technology over the other, and they wanted to create a level playing field for all participants.

Just because the competitors have access to the latest lidar and related navigation technology doesn’t mean they will have an easy time building capable autonomous systems. As Prucka explained, current lidar systems aren’t designed to scan environments fast enough for extreme speeds, which represents a significant hurdle.

“How do you process this data and make predictions under these kinds of conditions?” Prucka said.

It’s up to participants to decide how much they want their self-driving cars to push the limits when it comes to winning. Human drivers who notice that their cars are close to overheating, for example, may decide to take the risk and finish a race if there are just a few laps left. That’s a scenario that participants have to think about when developing their self-driving cars—the balance between safety and the desire to win. 

Prucka also said that the organizers are taking some safety precautions in case something goes awry. For instance, if a team loses a communication signal to their car for whatever reason, the organizers will perform an “executive shutdown.” 

Later next year at an unspecified date, competitors will test their deep-learning software in a virtual simulation race, said Ajei Gopal, the CEO of Ansys, which produces simulation software for engineers. While the organizers haven’t finalized what the virtual event will entail, “the intent is that the simulation would look close enough to the actual race.”

“We have never run a simulated race before,” Gopal said of the logistical challenges. 

If participants are able to complete the simulated race, they will proceed to the physical race. This portion of the contest will consist of multiple days of qualifying racing, followed by the big race on Oct. 23.

It’s unclear how the current coronavirus pandemic will alter plans if the deadly virus is still widespread next fall. 

“As of right now, we’re full speed ahead,” Peak said.