Will AI recreate the mind of a racing driver? Inject a line of code with the heat of competition, an instinct for a sure gap, a sense of the car's razor-edge limits, or precise positioning to frustrate rivals. Can it be injected?
The answer isn't immediately clear after last weekend's inaugural Autonomous Racing League event, billed as the first-ever open race between self-driving cars. In the grand final, one car spun on the first green flag lap, and the remaining three AI cars stopped behind it.
If this was what was meant by the series' claim to “redefine racing entertainment,” the audience was quick to pass judgment. As humans emerged from the pit lane to retrieve their cars, thousands of spectators in the stands stood up and quickly left.
“We are now exploring vast uncharted space.”
Earlier in the day, they had seen a watered-down version of the already highly publicized “man vs. machine” showdown, in which former Red Bull F1 driver Daniil Kvyat's duel with an AI racing car The laps turned out to be a series of demonstrations.
And that was preceded by a series of practice sessions and qualifying, where one AI car crashed into the barrier and another went straight into the second. For some teams, just getting around the circuit, let alone attempting racecraft, was difficult.
This could have been the event that showed once and for all whether wheel-to-wheel competition between AI cars could be an interesting alternative to human racers. But it had the distinct feel of a test session for technology that was far from ready. And it really was.
“We are currently exploring vast uncharted territory,” said Dr. Giovanni Pau, one of the team principals and technical director of Abu Dhabi's Technical Innovation Institute (TII). motorsport. “This technology is currently in its infancy. It's like a baby born yesterday taking its first steps.”
The Abu Dhabi-backed Autonomous Horse Racing League (A2RL) was announced exactly one year ago and has stuck to its ambitious schedule of holding its first race on Saturday despite setbacks that have slowed progress.
It started with 10 teams made up of the brightest minds in artificial intelligence, each with identical Super Formula-based vehicles equipped with 50 sensors, including cameras, radar, and lidar. These provide 15 TB of data per lap (the equivalent of 3 million songs), which must be processed and evaluated by a “stack” of AI software. The stack was trained by machine learning, and data from the simulator and actual laps by Daniil Kvyat were fed into the software, allowing it to identify patterns in track driving.
Each team then took that basic AI and developed it to improve the performance of their cars with the goal of competing against other teams on the track at last weekend's event. The idea is to race within a $2 million prize pool without any human input.
But the goal went beyond hosting a race series to accelerate the development of AI that would benefit the automotive industry and beyond. This was meant to be a new form of racing, attracting a younger generation of supporters who watched live streams via YouTube and his Twitch, and, of course, giving money back to sponsors. Additionally, thanks to free tickets offered on the series' website, the Yas Marina grandstand was filled with his 10,000 spectators.
There was already widespread skepticism heading into the race, and the Racing League, which has no further events scheduled this year, appears to have been unable to allay many doubts by proceeding with last weekend's race.
The vehicles only began track testing six weeks ago, and there was a wide range of performance due to varying amounts of experience between the team, which is made up of universities and technology companies. During Friday's test run, PoliMove, an Italian university team that has previously competed in other autonomous racing series, recorded lap times that hovered around two minutes, giving a preview of what's to come. The slowest runner was at least twice that.
“Our track time is very limited,” said Lawrence Walter, team principal of US-based A2RL racing team Code 19. motorsport In Abu Dhabi. “If you add up all the track time actually spent testing, [over the last two months] It's a number in hours. Therefore, it is extremely difficult to develop an AI race car driver from scratch in a short period of time, test it with limited testing time, and reach a competitive level. Many of our competitors have been racing autonomously for a long time. So for new teams coming in, the challenge is even greater. ”