To The Moon

Two years ago, $XPeng Inc.(XPEV)$ Xiaopeng started the expedition challenge of high-speed NGP assisted driving from Guangzhou to Beijing, and finally ended with an average bicycle NGP mileage of 2,930 kilometers and a total of 6,245 lane-changing overtakings.

Two years later, Xiaopeng once again launched the XNGP full-scene extreme challenge - passing through Guangzhou, Shenzhen, Huizhou, Hangzhou, Jiaxing, and finally arriving in Shanghai.

Unlike before, this time Xiaopeng's intelligent driving challenge is no longer limited to high-speed scenes, but crosses six cities in a row, regardless of urban areas or high-speed scenes and is a challenge to $Tesla Motors(TSLA)$ $Rivian Automotive, Inc.(RIVN)$ 

In this expedition challenge, Auto Heart also participated in the performance of the Shenzhen-Huizhou region. For example, in the face of the hell-level difficulty of intelligent driving in the Shenzhen region, whether it is the game of dense traffic or the connection between the overpass and the ground road, Xiaopeng XNGP's performance is very smooth.

Even if you enter the Huizhou area without a high-precision map, LCC-L can not only realize the experience of stopping at the red light and walking at the green light, but also determine whether it can quickly pass through the intersection at the yellow light according to the speed and intersection distance.

To put it simply, this is the first stage of Xiaopeng XNGP full-scene assisted driving. The biggest change is undoubtedly from the high dependence on high-precision maps in the past, and now mapless cities can still supplement their abilities through LCC-L after downgrade, and this version has also been upgraded to G9 and P7 at the end of March. I's Max user push.

"In the next stage (in the second half of the year), Xiaopeng will open the ability to automatically change lanes, overtake, and turn left and right in cities without high-precision maps." Wu Xinzhou, vice president of Xiaopeng Autopilot Center, revealed.

01. XNGP Phase 1: Move freely with pictures, and go straight without pictures.

From the perspective of functional definition, the full-scene assisted driving ability is the focus of competition among car companies. Xiaopeng calls it XNGP, that is, high-speed assisted driving + urban road assisted driving + parking lot navigation assisted driving full-scene point-to-point assisted driving function.

From the perspective of product completion, in the process of driving more than 120 kilometers across Shenzhen and Huizhou cities, whether it is high-speed or urban area, whether there is a picture or no picture, Xiaopeng XNGP has a leap in quality.

Let's start with the picture area.

"Zero takeover of the whole process" - this is Xiaopeng XNGP's 46-kilometer-long intelligent driving answer in Shenzhen urban area.

Normal urban road driving is no longer a problem for Xiaopeng XNGP. In the face of construction areas, cut in traffic lane changes, slow car detours, etc., the system can make early warning and flexible response performance.

For example, in the past, when encountering the forced congestion of the sidecar, Xiaopeng P5's reaction was to "hide when you see the car". When there was enough space left after the sidecar congestion, the system began to move forward slowly. This overly cautious strategy led to CNGP to bring certain security risks.

In contrast, as long as XNGP detects that the side car has the intention to add congestion, in addition to maintaining a safe distance from the side car to the left, the system will also actively initiate a lane change and choose a lane with less traffic to continue driving.

In fact, this is also in line with the driving style of human drivers, and the passage is completed more efficiently under the condition of ensuring safety.

After entering the highway section, we encountered a very challenging scene - rainstorm.

In stormy weather, rain, the water mist pulled by the front car, and the ground area of water will greatly affect the perception and recognition of the assisted driving camera. However, Xiaopeng XNGP can not only work more steadily, but also bring a decisive and smooth experience to the up and down ramps and lane changes.

In addition to the rainstorm section, the high-speed NGP under normal circumstances has shown a high maturity, including automatic lane change, upper and lower ramps and other functions. It is clean and neat, and there is no delay at all. There is no compromise in convenience and efficiency because of safety.

The "amazing" performance of XNGP is inseparable from the code refactoring of this system, that is, the high-speed NGP version 2.0.

Xiaopeng intelligent driving engineer revealed that the high-speed NGP 2.0 will continue to solve the long tail problem, including lane change ability, up/down ramp ability, traffic efficiency and other aspects.

Finally, let's take a look at the performance of XNGP without pictures.

Without relying on high-precision maps, Xiaopeng XNGP currently mainly implements LCC-L (road centering auxiliary enhanced version), which can bring traffic light recognition, start-stop, and laneless detours and other scenarios.

In short, you can also understand that this is the initial version of CNGP in a graphless state.

In addition to the scenes of automatic lane change, overtaking, and left and right rotation, LCC-L can achieve the same level as CNGP in terms of visual and detailed experience.

For example, for the SR visual interface, in the face of road information with no high-precision maps at all, LCC-L has been able to render quite finely, including curvature, dotted solid lines, zebra crossings, stop lane lines, vehicles, pedestrians and other traffic elements can be clearly presented in 3D models.

In a word, through the Shenzhen-Huizhou journey experience, we have been able to experience the full-scene assisted driving from the starting parking space to the end parking space, including the parking space, to the urban road, the highway/express road, to the city, and the final parking space.

This is the ability of Xiaopeng XNGP in the first stage, and the biggest highlight is undoubtedly the performance of "no graphics".

The engineer revealed that the current high-precision map only covers the main urban main roads, secondary roads and some branch roads. For example, the coverage rate of the CNGP high-precision map in Guangzhou is only about 30%. In addition to some temporary construction lane changes, these scenarios that need to be "taken in time" in the past can be supported through the branch. Supplementary LCC-L with no picture.

As for the expectations of the next stage, you can also imagine that this is an advanced version of CNGP in a mapless state - to realize lane change, overtake, and turn left and right.

$NIO Inc.(NIO)$  $Semiconductor Bull 3X Shares(SOXL)$ 

@Daily_Discussion @MillionaireTiger @Tiger_chat @TigerStars