After completing a test drive across China’s eastern coastal region, Xpeng Motors said on Wednesday that its driver assistance technology is the top performer in China, using a technology rejected by Elon Musk: high-definition maps.
At a press event in Beijing, Xpeng executives said its Navigation Guide Pilot (NGP) function, which enables primarily unassisted highway driving, surpassed Tesla’s Navigate on Autopilot (NoA) in several key metrics. Specifically, Xpeng said that it had achieved a lower rate of human driver intervention and a higher success rate for automatic lane changing, among others. The 3,600-kilometer (1,864 miles), eight-day road trip, which included members of the media, ended on Sunday.
The road trip included a fleet of 15 P7 sedans traveling a combined total of around 50,000 kilometers on highways and urban streets through major domestic cities including Beijing, Shanghai, and Guangzhou. Xpeng said it logged 0.71 disengagements per 100 kilometers. This means a human driver was forced to take control of the vehicle after traveling in autonomous mode for 140 kilometers on average. In the meantime, Xpeng claimed several Tesla vehicles in tests conducted by local media experienced 1.03 disengagements per 100 kilometers.
The Chinese EV maker also announced its latest version of NGP, scheduled to launch through an over-the-air update in the second quarter, resulted in a 94.4% success rate for lane changes versus Tesla’s 81.3%. Xpeng vehicles successfully self-navigated through tunnels 95.0% of the time compared with Tesla’s 41.8%. Huang Xin, a director at Xpeng Motors, called it “an overwhelming lead” (our translation).
”NGP completely exceeded Tesla’s NoA regarding all the metrics in our tests… and has become the most advanced driver-assist function for production models,“ (our translation) Huang said while calling out challenges from all of its competitors. Huang added that Xpeng will release all the data collected during the trip.
TechNode took one of the Xpeng sedans on a test drive from a hotel in Shanghai to a highway service zone in neighboring Suzhou city, sitting alongside the driver. During the 45-kilometer, 40-minute test ride, the vehicle drove primarily at around 120 kilometers per hour, navigated safely and responsively including changing lanes a number of times. However, at one point, the driver was required to take over the wheel when the vehicle passed an off-ramp on its right while being cut off by a car from the left.
In another test drive made by Chinese trade publication 42How, the P7 disengaged 19 times over 2,000 kilometers of autonomous highway driving compared with 22 driver interventions for a China-made Model 3 on the same route. The article said that Xpeng’s tech provided a better, more localized experience for Chinese customers, including a smoother drive when guiding its car from a highway on-ramp to off-ramp, and normal operation in tunnels or with heavy rain, which caused Tesla’s NoA to stop working.
Alibaba helps
So far, around 20% of owners of Xpeng’s P7, the company’s first premium model with the hardware necessary for offering advanced self-driving capabilities, have ordered its latest Xpilot 3.0 advanced driver-assist system (ADAS) featuring the NGP function, which launched in January. The Nio Pilot, which has been offering for almost three years, had a 50% take rate. More than 68% of Tesla buyers had reportedly opted in for its Autopilot software back in 2019.
READ MORE: Nio, Xpeng, Li Auto: your cheat sheet to China’s listed Tesla rivals
And yet, Xpeng is considered by many to be a big threat to Tesla in China where vehicle autonomy is concerned. Xpeng has boldly marketed itself as one of few companies capable of developing in-house the entire software architecture for AVs. The P7 currently remains the first and only production vehicle in the market equipped with Nvidia’s Xavier computer dedicated to highly autonomous driving, according to Xpeng’s vice president of autonomous driving Wu Xinzhou.
And now, the Alibaba-backed EV maker is stepping up its challenge against Tesla by working hand-in-hand with Alibaba’s map platform Amap, or AutoNavi. The company is confident that an elaborate, detailed map for real-time self-driving purposes would give it a leg up in luring increasingly savvy Chinese consumers, according to comments during the online press event. Xpeng attributed Amap’s latest high-definition map with providing navigational capabilities in adverse weather conditions or places with poor signal such as tunnels.
“Our vehicles can enter and exit highway ramps automatically and switch highways pretty much all by themselves, because most of the interconnections between highways are mapped by our partner AutoNavi. So we can have a seamless experience when you’re switching highways using NGP,” Wu said during an online conference in late January.
NGP could work properly in benign weather conditions, Wu added, and even under “medium to heavy rains” although it is designed to shut down and require human intervention when the windshield wipers are on the highest setting. Wu acknowledged there are also challenges in snow, which make it difficult for the vehicle’s sensors to detect road lane lines.
Tech dogma
The practice of using HD maps for AV navigation has long been criticized by Tesla’s Musk, partly because maintaining an constantly updated HD map was believed to be an arduous and costly effort. Musk in 2018 publicly stated that dependency on HD maps would cause an AV to fail when real world changes are not reflected on the map. Tesla’s vehicles, he said, have sufficient sensors and processors to drive themselves.
Tesla did not respond to TechNode’s request for comment.
However, most other automakers and AV companies including Waymo and GM Cruise, rely on a suite of hardware stacks comprised of cameras, radar, Lidar, and HD maps—usually viewed as “another sensor.” Xpeng is currently the only car company incorporating Amap’s latest map technologies for on-board navigation, a partnership which Wei Dong, a general manager of Amap, commented requires an automaker have a strong proprietary capability in software development, since map data will be aggregated with sensor data to give AVs a sense of their surroundings.
“We do a very careful checking between what the cameras see and what the map is telling you pretty much all the time. And whenever there is a difference, the system will send a warning to the driver and sometimes just downgrade the AV functionality to make sure it’s safe,” Wu told TechNode.