I expect Tesla’s auto driving software will hold sway in the US & EU. But what about the China market place?
XPeng Inc. (“XPeng” or the “Company”, NYSE: XPEV), a leading Chinese smart electric vehicle (“Smart EV”) company, today released the daily results for the first six days of China’s longest real highway autonomous driving challenge by mass-produced vehicles. These figures set some of the most successful autonomous driving performance records to date over long distance in real road conditions in China.
Under the control of XPeng’s Navigation Guided Pilot (NGP) autonomous driving assistance, the frequency of human driver intervention was as low as 0.65 times per 100 kilometers, and was under one time per 100 kilometers in five out of the six days from 19 to 24 March on highways in China. The drive has covered some of the busiest and most complex highway sections along the country’s eastern costal corridor.
The success rate for lane changing and overtaking has been between 86.05% to 97.91% during the 6-day period. Highway ramp entering and existing success rate ranged from 83.76% to 96.93% since the P7 fleet set off from Guangzhou on 19 March, passing through Shanghai and arriving in Qingdao on 24 March. Under the NGP control, tunnel pass-through success rate was as high as 100% during the same period.
The NGP delivered stable performance in rainstorm conditions and on highways in mountain areas during the 6-day expedition, showcasing the robustness and reliability of its functionality for complex driving scenarios in China.
XNGP replaces Xpeng’s Xpilot ADAS system. The company said XNGP will roll out later this year in certain cities in China and it will be available with the G9 Max, its top-spec sport utility vehicle. XNGP is Xpeng’s most advanced ADAS to date.
Features of XNGP will be rolled out over the next two years. City NGP, which stands for navigation-guided pilot, will be rolled out starting this year and into the first half of 2023 in certain Chinese cities. City NGP allows the car to semi-autonomously navigate complex urban environments with features such as lane switching.
In 2024, Xpeng is aiming to have semi-autonomous driving features available for all major cities in China.
Tesla will be buying a number of those cars to dissect them for their hardware/software. Tesla’s FSD is nowhere near that good, but that also means we do not know how the test results were obtained (was a third party doing the testing/rating? Or…???).
One has to wonder what “failure” looks like when you are changing lanes, overtaking, and entering/existing the highway. Even a 2% failure rate would mean a hell of a lot of failure on the highway all day long.
tunnel pass-through success rate was as high as 100% during the same period.
o.0 chuckle What does ANY failure at tunnel pass-through look like? I am envisioning scenes from Wile E. Coyote.
From the article:
A live webcast will be hosted by Tom Moloughney, a veteran EV advocate and Mark Andrews, an experienced auto journalist who drove a P7 for one day of the event, to analyze the NGP’s performance in this cross-country autonomous driving expedition, along with footage of his drive.
Heh, until the government can mandate a recall on human failure, I am not sure it matters (though I am very confident it is significantly less than 2% for merging with traffic).
A lot depends on the definition of a failure.
Failure could mean a crash occurs.
It could mean you start to attempt a maneuver (like a lane change) and cancel because you can’t merge so you just stay in your lane.
It could also mean that you complete the merge but others cars have to slow or take other action to prevent an accident.
In my Tesla the most common failure I see is a lane change needs to take place in heavy freeway traffic and it just can’t do it. Human drivers are still much better via more aggressive driving…such as putting on a turn signal and moving over slightly in your lane when there is a possible opening and the car beside and behind yields a bit to let you in. If that car speeds up instead you try the next opening.
The thing is … humans are also still much better with sedate driving. My Tesla on autopilot will speed up rapidly to reach the set speed, and will slow down very abruptly when approaching slowed or stopped traffic in the lane ahead. It’s also so annoying how it will hard brake at various times - for example, a few days ago, someone was making a left from the opposing traffic ahead of me, way ahead of me, but just AFTER the other car cleared my lane, my Tesla slammed on the brakes, and remained moving very slowly LONG AFTER that car was gone.
A human (me!) can tell that if there is a row of cars ahead, perhaps at a light, or perhaps simple highway traffic, all with brake lights on and not moving, or moving very slowly, that they (I) need to slow down and perhaps stop. So I will gently slow down and approach the back of the row of cars. Meanwhile, most cars, including Tesla doesn’t even “look” at brake lights (as far as I am aware).
A human can see or know, that the road ahead has an steep incline immediately followed by a steep decline, so the human will allow the car to slow a bit going up the hill, and then allow the car to use the momentum going down the hill to speed up again. Meanwhile, all cruise controls (even the fancy ones like Tesla) will zoom up the hill trying to maintain the set speed at any cost (lots of fuel or battery power), only to have to slow on the way down the other side, sometimes braking (using regenerative braking).
Humans still do a lot of driving tasks better. But automated systems do other driving tasks better, mainly the boring stay-in-lane, keep speed, keep distance, highway driving.
But the delays I encounter regarding braking and warnings is a sure sign that NOT A SINGLE car sold today will EVER be fully self driving. For the simple reason that none of them have sufficient processing power to do all tasks all the time.
The dream will be a failure until we move to mutually communicating and cooperating vehicles that would make driverless driving actually real, but the far bigger benefit will be in making driving far more efficient both in use of energy (far far less rapid braking and rapid accelerations) and a radical reduction in number of paved lanes required.
Yep. Pretty much entirely true. We (the USA) were well on our way to having vehicles communicate with one another, but feckless auto companies combined with an institutionally corrupt FCC killed it in its tracks. Ironically, most of the bandwidth that was allocated to this was re-allocated by the FCC to essentially provide more wireless access to cute cat videos.
Why? An autonomous car needs to be aware of (and know how to react to) an untold number of physical objects in its surroundings other than other cars. None of which will be communicating with the car. So any car that’s capable of operating autonomously will, by necessity, have the ability to properly respond to other vehicles also, whether they are mute or not. What is the marginal benefit of having them communicate with each other directly?
In a situation where two (or more) vehicles encounter a situation where one or both need to take evasive action, wouldn’t it be good if the vehicles in the situation were able to communicate in a split second → Vehicle A goes left, Vehicle B goes right - to avoid a collision?
ie: what would happen if both autonomous driver systems told the steering mechanisms to steer into each other’s line of travel?
I can see advantages with driverless cars to be able to communicate with other vehicles around them.
It would - but that’s a very unusual edge case. With human drivers, of course, they might not react in time to avoid driving into each other. With an AV driver, presumably, the reaction times are much faster. So if Car A starts moving into Car B’s line of travel to avoid something, Car B immediately knows this whether Car A is communicating or mute. IOW, the mere act of Car A starting its movement into Car B’s line of travel to avoid an obstacle communicates to Car B nearly all of the information Car B needs in order to avoid the collision with Car A. Only in very rare circumstances would having communication help at all - where a tiny fraction of a second of more advance notice would matter. That’s probably not worth it, TBH.
Remember, communication doesn’t necessarily mean collaborative decision-making. Each car is still going to have to decide what to do on its own. And since the cars can see each other pretty well without them “speaking” to each other, and can react to each other much much faster than a human could even when mute, there’s just not a lot of benefit here.
If all cars on the road have the ability to communicate with each other, I can see the advantage. What about all of the cars that don’t have that ability? For the foreseeable future they will be the majority on the road.
Yes, but that’s always a problem when adopting new technology. It will take probably 10 years from the start until half the cars have it, and then another 10 years until “all” (the vast majority) have it. Just like rear view cameras. Rear view cameras were mandated in 2018 model year, but were popular among high-end models for a few years before that. Now perhaps 25-35% of cars on the road have them. It’ll take another 7 or 10 years until the vast majority have them.
The advantage might be in planning ahead rather than reacting to situations.
Airlines have TCAS (Traffic and Collision Avoidance System) that lets planes know they are getting close and telling each what to do to avoid a collision. One would think that a similar system would be helpful in cars as well.
Well, planes can’t “see” other planes. They’re not autonomous. They’re not driving themselves.
We’re talking about alternative systems of fully autonomous cars. These cars can already “see” each other, whether they communicate or are mute. They have to be. They’ll be able to see lightpoles and curbs and crosswalks and pedestrians and traffic cones and literally an uncountable number of other “mute” objects - so they’ll be able to see each other.
Each of these cars will need to be able to avoid a collision with a “mute” car that might swerve into its lane for the next decade or two - because it will be driving mostly with mute cars for around that time. That type of vehicle won’t need a transmission from the other car to know where it is and whether it’s getting close. It almost certainly won’t need as much advance warning as a human driver, either, in order to avoid a collision.
Just a couple of interesting data points. (There are probably numerous sources for this type of data )
The National Highway Safety Administration reports that 62 percent of all fatal California car accidents were single vehicle accidents
These are probably mostly eliminated with autonomous cars and do not require any V2V communication.
Rear-end collisions are by far the most common type of car accident.
Note that this is accidents, not deaths. Probably eliminated if ~all cars have V2V communication but will take decades to get to this point
The edge case of two cars taking conflicting evasive action doesn’t show up in any top accident lists I could find.
Human drivers do OK without communicating beyond the car’s signals. uTube videos are starting to show up with the latest FSD 11.3.1 beta which combines city and highway driving (full stack).
The improvements are quite stunning, FSD learns quickly. That is not to deny that car to car communications (IoT) would be a good idea. In one presentation they mentioned that cars are reporting local conditions to Tesla which makes that info available to cars on a geo location basis. That is a step in the right direction [sorry, no link to this last bit].
Officially certified vehicles loaded with Level 3 autonomous driving technology in the global market are the Mercedes-Benz S-Class and electric sedan EQS, and the Honda Legend. Drive Pilot, Mercedes-Benz’s level 3 autonomous driving technology, empowers cars to run at speeds of less than 60 kph.
