Every automaker and their cousin is going to lead us into an autonomous driving future … according to their press departments. In actuality, I think there are only a few automakers that have deep in-house teams focused on impressive internal innovation to get to that eventual goal. In the shorter term, that kind of leadership also allows them to have the best ADAS (advanced driver assistance system) features. Aside from Tesla, the top company I’ve been keeping an eye on in this field is XPeng. I know they have a robust internal team focused on full-stack development of autonomous driving tech and I know they are taking an interesting approach that integrates more types of hardware than Tesla (lidar, radar, and cameras instead of just cameras). On the software side, I would guess that they have more software developers focused on this than anyone but Tesla. In short, I think it’s one of the top companies that could potentially innovate itself into a pole position in this field — temporarily if not even in the long term. I’m not predicting that (I’m not making such a prediction for any automaker, autonomous driving startup, or automotive supplier), but it’s one clear possibility.
With that context in mind, I found it interesting having a look at how the CEO of XPeng, He Xiaopeng, responded to the most advanced version of XPeng’s “City NGP” (City Navigated Guided Pilot) on a recent test drive of the tech. Watching the video, I was even more impressed to see that he was very open and critical about weaknesses he saw in the tech, while also of course praising some of its more advanced capabilities. Having a Tesla FSD Beta in my own car, I know quite well some of the biggest on-road challenges to this kind of tech, and you can see He Xiaopeng point out some of these exact challenges in the video. He highlights fairly nuanced issues that keep the autonomous driving software from being as smooth and relaxing as a normal driver can be. Watch the video below and then I’ll come back to some of these items more specifically.
Just looking at this short video of City NGP’s capabilities, the tech looks not that much different from Tesla Autopilot. It appears to be excellent at lane keeping and traffic-aware cruise control (aka adaptive cruise control), and it seems highly adept at changing lanes automatically, but with some little niggles like being too passive at times or not recognizing the context of a jam -packed traffic jam in which other cars behave differently and won’t give you much space to get into an opening. Though, the VP of XPeng’s Autonomous Driving Center, Wu Xinzhou, notes in the video that a future version of the tech will take note of such traffic situations and ask the driver to assist in changing lanes in those environments. It’s not a perfect robotaxi approach to this challenge, but it seems like the right one for this stage of the technology.
He Xiaopeng also notes a moment where a car somewhat aggressively entering his lane is noticed by the XPeng City NGP suite and the software very appropriately provides some extra space for that car to enter. That level of seemingly natural human-like adaptation is impressive, and it shows how the XPeng team is working to make City NGP as smooth and predictable as possible.
Anyone who has followed my commentary and review of Tesla FSD Beta knows that one of my main concerns and top reasons for disengagement is potholes. No one wants to hit potholes, but the system isn’t yet capable of handling them. He Xiaopeng specifically asks about potholes and how the City NGP system handles them, and Wu Xinzhou indicates that it can’t yet recognize and avoid them well, but that the team is working on that and the next version of the tech will “most certainly “be able to recognize them. Personally, I find it refreshing that they are willing to share these kinds of existing weaknesses publicly like they are doing (not every company would highlight these things — and certainly not through a CEO test drive video). That kind of authenticity and honesty is appreciated, and I think not just by me. I also find it encouraging that they are focused on solving the pothole problem. I’m extremely curious to see how well they do that in the next version of their semi-autonomous driving tech.
Wu Xinzhou also spends a little time explaining how lidar helps the XPeng team to handle certain things, like heavy rain. Again, this is an area where Tesla’s system can struggle a bit due to being 100% vision based and the visibility challenges in such rain. I would love to compare the two systems side by side in similar rainy conditions! Maybe XPeng could lend us a vehicle for that someday. 😉
Wu Xinzhou also talks about ways that lidar helps the company to spot and avoid static obstacles. I’m not sure if that is purposely throwing light shade on Tesla’s approach, but it is a known problem Tesla vehicles have faced from not using lidar. Naturally, Tesla is working to solve for that problem in other ways, but it seems noteworthy that XPeng is explicitly highlighting that this is one reason they see lidar as being critical.
For the computer geeks out there, the most impressive note of all may be about the computing power XPeng is using for all of this. “We are the only company in the world that managed to develop city-level ADAS functions based on only 20–30 TOPs of computing power,” Wu Xinzhou states. But rather than simply pat his own back at that truly impressive achievement, he immediately jumps into some self-critique. “But we also have limitations. For certain specific situations, the computing capabilities are simply not enough.” He then talks about some of the edge cases they are working on, such as people lying down, people in wheelchairs, and people squatting.
Overall, I thoroughly enjoyed what seemed to be an open, candid take on where XPeng’s semi-autonomous driving tech sits at the moment. I strongly encourage you to watch it. If any additional thoughts come up while watching the video, chime in down in the comments below and let us know.
Appreciate CleanTechnica’s originality and cleantech news coverage? Consider becoming a CleanTechnica Member, Supporter, Technician, or Ambassador — or a patron on Patreon.