2020 XPeng 1024 Intelligent Day
After experiencing XPeng NGP engineering edition, my colleague 63 told me that his XPeng P7’s assisted driving could beat my car.
As a Tesla and Nio owner, my feeling can be described with the word “hehe”. At that time, my impression of XPeng was just “Attention: Back up”.
With this mindset, I participated in XPeng’s NGP Beta version experience activity in Guangzhou on January 12, and I just want to know what’s so special about the NGP that my colleague praised.
The most intuitive change in this NGP Beta version is the UI design and human-machine interaction.
The NGP has completely redesigned the navigation page of the entire central control screen into a visualized page for assisted driving, and it is 3D modeled by combining high-precision map data.
Not only does it display lane lines, cones, and other traffic participants on the screen, but it also adds information such as distant buildings, mountains, and rivers to the central control screen, even pedestrians on the side of the road.
However, this is not just to look good, but XPeng knows very well that its perception system based on 13 cameras and high-precision maps determines the ability boundary of the NGP.
From an industry perspective, this is currently the system that does visualized assisted driving most comprehensively. Integrating high-precision map data with assisted driving allows the NGP to pass through scenarios where some lane lines are not clear. In such scenarios, Tesla NoA will probably perform “drawing dragons” within the lane lines.
However, I have a question. The NGP visual page of the central control is already full and cool. Can the dashboard also be arranged?
In addition to discussing the NGP experience, I have always believed that for auxiliary driving, the experience of the two parts of lane changing logic and entering/exiting ramps directly determines the frequency of use by users.
Let’s first look at the lane-changing logic. I believe that the lane-changing of auxiliary driving is mainly to pursue higher traffic efficiency under the premise of ensuring safety. The two aspects that affect the experience are the triggering conditions for lane-changing and the execution efficiency of lane-changing.
Speaking of the triggering conditions for lane-changing, the NGP of XPeng can perform lane-changing even on highways with a speed limit of 120km/h, even if the speed of the front vehicle is maintained at 117km/h. Actually, regarding this setting, for me, I think it is good, but my requirements are not high.
There are three reasons. Firstly, this logic is not too friendly to the range of my Nio. Secondly, I think that the highest priority of auxiliary driving is still to reduce driving labor intensity, and traffic efficiency ranks second. Therefore, finally, the overly active lane-changing logic is not good for the experience on city elevated roads.
Although the NGP’s lane-changing strategy is very active, in the scene of the following video, the NGP was originally going to overtake by changing lanes to the left. After turning on the turn signal, the vehicle in the front left suddenly slowed down, and the NGP quickly cancelled the lane-changing command. This is great, and it’s more “humanized.”
Moving on to the lane change efficiency, NGP performs exceptionally well in this aspect. After attempting to manually activate the turn signal a few times, NGP’s lane changing response time is very fast. Let me provide a small example for everyone to compare with the reaction time of traditional fuel car transmissions.
The slowest response time is the type of transmission that is on the brink of elimination, the 4AT transmission. The majority of these driver assistance features are bundled by suppliers. Subjectively speaking, “I’m faster than it”. After the turn signal is activated, the driver assistance chip takes several rounds before it can make the corresponding lane change motion.
The second type is like a certain brand’s 8AT transmission, which is usually smooth, but when it comes to full throttle situations, it always can’t find the right gear. This is like some car models on the market that are fine with lane changing under normal circumstances, but when the traffic volume is slightly higher, they flop. If manufacturers ignore the inherent hardware limitations and only talk about conservative strategies, this kind of driver assistance can be a bit “rogue”.
The third type is the dual-clutch transmission. Fast and precise. I believe that NGP belongs to this category. As for which specific dual-clutch brand NGP is benchmarking against, it’s like this: It’s almost there except for that little bit before the kind that can jump for an afternoon without getting hot.
And NGP’s strength is not only in the better lane change experience on open roads.
The key point is that this system has a “brain”. For example, when encountering a scenario where the vehicle will exit the ramp and needs to switch to the right lane, but there are vehicles parallel on the right lane, at this time, NGP will attempt to accelerate or decelerate and merge in front of or behind the parallel vehicle. My NIO car just basically “raises the white flag” in this kind of scenario.
You can’t have your cake and eat it too
Now let’s talk about the experience of NGP entering and exiting the ramp. Two kilometers before the ramp, NGP initiates the lane change, leaving enough time and distance redundancy. Because at this time, the priority of smoothly exiting the ramp must be higher than the traffic efficiency, there is no problem. Coupled with excellent lane changing logic, NGP’s preparation work for entering the ramp is very smooth.
After entering the ramp, with the real-time adjustment of the speed based on the curvature data provided by high-precision maps, it is smooth to pass through the ramp, and the experiences are all great.
When leaving the ramp, NGP only increased the speed to 60 km/h within the ramp, and accelerated after merging onto the first main road on the far right, which is acceptable. As always, under the current scenario, the priority of smoothly entering and leaving the ramp is the highest.
However, based on the comprehensive experiences, there are still some areas that NGP can improve, for example, limiting the speed to 60 km/h 200 meters before the ramp, which may cause embarrassment when a large truck passes by faster than the car. Although the staff of XPeng on-site has arranged the requirements with high efficiency.
In addition, in some ramps with very large curvature, NGP may be a bit too aggressive to maintain a speed above 60 km/h when entering the curve, which can make the driver feel oppressed. In the video, I took control of the steering directly, which was the only time I took over during the whole journey. In this regard, my NIO NOP has gained an advantage.
Regarding the strategy of changing lanes when entering or exiting a ramp, there is one more point I want to make. In the video, the NGP decelerated and changed lanes into a gap between two cars as it approached the exit ramp. This action in itself was not a problem, but as the lane change was executed past the solid line marking, the operation was cancelled and returned to the original lane.
In fact, there have been heated discussions about whether to cross the solid line inside XPeng Motors. My understanding is that from a safety perspective, I support crossing the line. This is because the perception system has determined that the right lane is safe, and the risks are minimal when changing lanes.
But if we were to withdraw, there would be some uncertainty. This is because the millimeter-wave radar behind our vehicle is usually used to perceive the diagonal rear position, not directly behind. The uncertain risk is that when the car behind sees the car in front changing lanes and giving up the lane line, the driver can accelerate. However, when the front car cancels the action halfway through the lane change, this unexpected driving route is a risk to the driver of the rear vehicle.
Although due to time constraints, I did not experience the NGP in congested or rainy weather conditions, in terms of clear weather scenarios, the NGP’s better continuity performance and behavior closer to that of human drivers have given me a deep impression.
In the era when everyone is still using advanced driving assistance, the NGP’s algorithmic logic is a lesson for the industry. As my colleague from 63 said, it performs better in high-speed scenarios than my Tesla and NIO experiences, and the NGP is also getting closer to me who just got my driver’s license a decade ago.
Although the current beta version still cannot achieve the goal of XPeng’s classmate, which is to travel 2000 kilometers without touching the steering wheel, we have seen that XPeng has injected algorithms with more “human-like” logic into this advanced driving system. We also look forward to the NGP final version making XPeng’s classmate’s goal a reality.
This article is a translation by ChatGPT of a Chinese report from 42HOW. If you have any questions about it, please email email@example.com.