Over the past few years, we’ve seen computerized systems meddle with many things. Reasonable expectations of privacy, your ability to do your job without wading through whichever program management’s foisted on you this quarter, even the concept of reality judging by how many people seem to fall for AI-generated images. Now they appear to be coming for the art of driving fast. The first officially invigilated autonomous Nürburgring Nordschleife lap record has been set, and it’s a bit underwhelming.
The weapon of choice for this record-setting run was the Xiaomi YU7 GT with the optional track package, an electron-fed 1,003-horsepower super-SUV from a brand that briefly held the Nürburgring EV record. We’re talking epic figures here: zero-to-60 mph in under three seconds, a top speed of 186 MPH, and a human-piloted Nordschleife time of 7:22.755. With hardware like that, we should be in for a reasonably quick autonomous lap, right?
The good news is that it did manage to go ’round without crashing despite nobody sitting in the driver’s seat. The bad news is that its record-setting time came out to 10 minutes and 29.483 seconds. Given the Nordschleife’s 12.944-mile length, a lap time of 10:29.483 works out to an average speed of 74.09 MPH. That’s 119.23 km/h in metric.
Granted, the autonomous YU7 GT did manage to beat some stuff. It’s quicker than a Ligier or a Trabant or a Mercedes-Benz G55 AMG. However, it’s not slower than perhaps the most culturally-known Nürburgring lap in a slow-ish car. Remember on “Top Gear” when Jeremy Clarkson ran a 9:59 lap in a diesel Jaguar S-Type? That wasn’t actually a full-length lap, but instead a “bridge-to-gantry” lap from the first bridge after the join to the last gantry before the exit on open days. If we’re ballparking based on the onboard video, the autonomous YU7 GT seems to have set roughly a 9:57 bridge-to-gantry time. That’s not hugely impressive considering this is a massively quick EV on one of the world’s great power tracks boasting a 1.3-mile straight and the cultural benchmark was a diesel executive sedan with a zero-to-60 mph time of 8.5 seconds and 22-year-old tire technology.
Why so slow? The weather, for one. We can see from droplets falling on the windscreen in the onboard video that this record attempt was made in slightly damp conditions, which aren’t optimal for maximum pace. However, the autonomous Xiaomi seems excessively cautious at times. Through Hatzenbach, we see a minimum speed of just 63 km/h, or 39 MPH. The YU7 GT never exceeds 150 km/h through Flugplatz, or in the flat-out left at Kottenborn. A minimum corner speed of 60 km/h in Aremberg is on the slow side, even in much wetter weather. Oh, and the car maxes out at a mere 140 km/h through Fuchsröhre, which is a fast descending section with the left-hander at the bottom being both on-camber and a compression. Indeed, the autonomous Xiaomi continues in this excessively cautious style throughout the entire lap. No wonder it couldn’t break the 10-minute mark.
Granted, dialling things back to a cautious pace isn’t necessarily a concern on its own, but one major detail here suggests that Xiaomi’s circuit-ready system isn’t quite ready for prime time. When water falls on a road course, it typically does two things: Pool more in some places than others, and make the rubbered-in dry racing line incredibly slippery. To compensate, keen drivers usually adopt a wet line, avoiding the slickest parts of the driving surface. In contrast, the YU7 GT appears to have driven the dry line the entire way around the Nürburgring, slowing to a crawl on the rubbered-in part of say, the entry to Wehrseifen instead of taking a late apex in order to brake later. On principle, so long as an autonomous vehicle stays within its designated lane, avoiding the slipperiest parts of the lane seems more proactive than compensating with speed reduction and other corrections, right?
By now, you might be wondering what the point is of autonomous Nürburging laps. As much as I believe in the synchronicity of human and machine, allow me to make an argument. For an autonomous car to be safer than the typical human driver, one of the things it must do is exhibit particularly good car control. The ability to make a quick lane change while maintaining control to avoid something popping into the road around a blind corner, or over a blind crest, or in any other situation where systems like Lidar and radar couldn’t detect an obstacle well in advance and the distance doesn’t quite exist to slam on the brakes and avoid a collision that way. What better place to hone autonomous car control than on a notoriously fast, bumpy, hilly track?
This isn’t the last we’ve seen of Xiaomi’s autonomous systems. Or any autonomous systems, for that matter. Consider this slower-than-a-Mk4 Volkswagen Polo GTI lap time the benchmark, because autonomous driving systems will return to the Nürburgring showcasing higher levels of control. They still have a long way to go in order to match the pace of a racing driver, but autonomous driving systems that work well on track should be safer under pressure on a road near you.
Top graphic image: Xiaomi
I’m not convinced of this. A closed track is probably one of the easiest scenarios to handle for an autonomous car. Hell, every racing game in existence has AI that can handle driving around a track. The thing that makes autonomous driving difficult is the edge cases – people/animals/objects entering the road unexpectedly, weather that fouls the sensors, etc. Cruising a long, unbroken stretch of pavement is not the hard part.
Frankly more surprised this hasn’t happened sooner.
Looks like the machines aren’t quite ready to take over. I’d love to see that car shit itself trying to deal with other cars on a live track.
Invigilated? I had to look that one up. “Invigilate” is almost never used in American English; I’d guess 99% of American English speakers wouldn’t know what it means but I might be wrang (archaic form of wrong).
Do you really think that the track would be considered a) “rubbered in” when it was reserved for the manufacturer and probably only been driven on by a single or at most a couple of cars, and b) wet enough for a wet line to make a huge difference in pace? I think most of the caution we see here is due to limits set by the manufacturer and current limitations of the system. I think it is still very impressive to see an autonomous car at what would be Touristenfahrten speeds for most people.
While I understand this from a development perspective, having autonomous cars drive on race tracks sounds about as fun as outsourcing creativity to AI.
Someone needs to organize an autonomous car race. Things I’d like to see, hold a series and some real road race tracks (no ovals etc). I might even pay money to watch it; from a safe well protected distance of course.
As the famous old racing quote goes “To finish first, first you have to finish!”
A crash would be embarrassing and you need to establish a baseline before finding limits.
I’d like to raise the same episode as Clarkson’s Jaguar with Hammonds Ford Transit van at 10 minutes 8 seconds. So this Supercar can’t beat a 2014 Ford Transit van.
The car looks like it is only processing the portions of the track in view of the cameras. It’s been 10+ years since I’ve played Grand Turismo and driven The Nurburgring, but learning what lines to take flat out around the blind corners was the hardest part of getting decent lap times. The computer seems to be keeping the speeds down in case it needs to stop for something just out of the camera’s view. Remove that safeguard and I bet we would see a blazing fast lap… or a fireball!
It surely can do better than this. This lap time is most likely achievable using cams and gears.
I am sure this will be eventually bettered massively and probably sooner rather than later it will get close to human drivers, at least in ideal conditions.
As they customize and prioritize and close the track
“…slightly damp conditions…”
I refer to these conditions as mildly moist because of Top Gear.
So it added 4 seconds?
At least it didn’t crash into someone’s house and kill the Grandma inside.
https://abcnews.com/US/tesla-allegedly-autopilot-mode-crashes-texas-house-woman/story?id=134062374
One step closer to a real Lightning McQueen…
We can only hope not considering Torch’s Homunculus Theory.
https://www.theautopian.com/how-the-motorhomes-and-campers-in-pixars-cars-universe-hint-at-a-disturbing-reality/
https://www.jalopnik.com/this-disturbing-theory-explains-pixars-cars-1791834045/
Perhaps this is how it begins!
Well, when Pixar’s Cars enters the public domain in 2102 (!!!), maybe someone will do a horror movie based on this concept.
Why wait?
https://uslawexplained.com/parody