Turn and make a U-turn in one go, efficiently avoid motor vehicles, non-motorized vehicles and pedestrians at complex junctions, avoid congested lanes and choose idle lanes wisely. Recently, Zhiji released an actual video of the IM AD city NOA in Lujiazui, Shanghai, which stunned the audience as soon as it was released.
Figure: measured route
On the core and most prosperous sections of Shanghai, the complex and changeable road conditions, dense flow of people, and vehicles parked on the side of the road are all hell difficulty for Zhiji, who has just embarked on the intelligent driving road for two years. But Zhiji's performance is very bright, no less than professional drivers: no frequent frustration, comfortable start and brake, smooth in and out of corners. It can be said that Zhiji has refreshed the landing speed of smart drivers in the industry, taking two years to complete nine years of smart driving of the industry's leading enterprises, and is leading the arrival of a new era of urban NOA.
With a soft car and a fast start, Zhiji made a "city NOA that is more like an old driver".
NOA (Navigate On Autopilot) refers to automatic navigation and driving, which is driven by navigation maps. Vehicles can automatically change lanes, overtake, start and stop, enter and exit ramps and other operations in highway conditions. "No car is not intelligent", and NOA is the most important application direction of automobile intelligence-the landing of autopilot.
Previously, major manufacturers have cut in from high-speed NOA, because its application scene is relatively fixed, and the road conditions are relatively less complex and less difficult; now it has transitioned to the city NOA, which can meet the driving needs of most car owners in core cities, and 2023 is also regarded by the industry as "the first year of urban NOA". Urban NOA has also become one of the most important functions for many car owners when buying new energy models.
Picture: 106° iMax wide field of view
According to the Tiber Research Institute, the number of models carrying NOA is expected to exceed 4 million by 2025, and the permeability is expected to increase from 12% in 2023 to 32%. NOA technology will undoubtedly become the mainstream trend of the automobile industry in the future.
However, the transition from high-speed NOA to urban NOA is not simple, because urban road conditions are far more complex than highways, and there are all kinds of emergencies, which require a high level of intelligence of automobile companies. The urban NOA of many automobile companies is "loud thunder, little rain". At present, Huawei and XPeng Inc. are the only players who really push urban NOA functions to users on a large scale in China.
Now, NOA, the city of Zhiji, has contributed silkiness, fluency and maturity on the hell-difficult Lujiazui road in Shanghai.
The driving performance is comparable to that of the "old driver", which is not inferior to that of Huawei and XPeng Inc., which undoubtedly brings greater pressure to other car companies.
Judging from the measured video of driving on nearly 10 kilometers of roads, the road condition is very complex, pedestrians, electric cars, and cars are dense, some roads are narrow and crowded, and there are many situations that need to make route judgments. however, Zhiji contributed extremely clever detour, road choice behavior, but also successfully achieved a U-turn, the response under difficult road conditions is even better than people.
For example, in a complex left-turn scene, the system first avoids the two straight vehicles in the opposite direction, and at the same time identifies the non-motor vehicles blocked by the road guardrail, waiting for the right time to ensure safe and smooth passage. The excellent performance of multi-target recognition and shelter recognition is due to Zhiji's leading visual perception solution, which makes it have the six-fold coverage perception ability of 360-degree non-blind area.
Figure: measured site
Picture: side blind area image + BSD side blind area light effect reminder
On a path where there is no lane in the middle, the roadside is full of vehicles and pedestrians are constantly shuttling, the system can calmly bypass roadside parking vehicles and opposite cars, and can automatically brake and avoid pedestrians, which is quite amazing. When there is a sudden entry of the battery car behind the vehicle, the system can also make an alarm prompt in time, actively slow down and evade. In addition, the system can find congested lanes and automatically choose a more idle lane next to it. The strong data-driven prediction ability can better judge the intention of other vehicles and improve the success rate of lane change, while the regulation and control algorithm can achieve lane-changing capacity similar to human through a large number of simulation data iteration, and effectively improve traffic efficiency.
Figure: measured site
Figure: measured site
At the end of the video, Zhiji also announced the unprotected U-turn scene at the crossroads for the first time, and the system perfectly planned a reasonable U-turn curve while observing the traffic environment in front and rear at the same time. This puts forward extremely high requirements for the ability of front and back detection, lane identification and steering planning.
Figure: measured site
Image: narrow road automatically triggers panoramic image
Just a few minutes of video, from the reaction speed of driving, route planning, vehicle avoidance, lane change and even driving details, Zhiji's performance is very close to human driving, better than the first-class level in the shoulder industry. It is understood that the Planning algorithm based on D.L.P. Deep learning model enables Zhiji NOA to wriggle and follow the car like a human being in a traffic jam, avoid frequent frustrations, start quickly and timely, follow the car softly, start quickly, accelerate smoothly, and have a high humanity. The appearance of NOA, a city of Zhiji, is like a stone that ripples in the pool, proclaiming the strength of the number one player.
What is even more difficult to imagine is that Zhiji completed all this in only two years.
Fast and good, cool techs helped Zhiji to enter the first echelon quickly.
In November 2021, Zhiji released its corporate innovation strategy, and in April this year, Zhiji released the industry's first D.L.P. Artificial intelligence model. On August 16th, Zhiji released a roadmap for the landing of smart driving products for the next 3-5 years, and announced that it would officially launch the public test of "IM AD City NOA" in October this year. Looking back on the development of Zhiji's intelligent driving, Zhiji has completed the 9-year road of intelligent driving in a short period of less than 2 years.
NOA's broad business prospects, so that this track has never been calm, and even the competition is still further upgrading, as a "latecomer" Zhiji, how to achieve in just 2 years to catch up with the head of the industry?
First of all, there are solid underlying technological advantages. The core competence of joint development with Momenta, with more efficient architecture, rapidly evolving algorithms and higher utilization of generated data, is an important factor driving the evolution of IM AD intelligent drivers.
Take the trend of the whole industry to go to high-precision maps as an example. Gaojing maps are described as the "crutches" of autopilot, but because of their high cost, long renewal cycle, and strong dependence on policies, major manufacturers have turned to "light maps" one after another. For Zhiji, its structure is lighter, the adaptability is more efficient, the computing requirement is reduced by 90%, and the running efficiency of the model is greatly increased by 500%. Based on this, limited computing power can also achieve global visual fusion, and the scheme of "single lidar + single Orin chip" can also cover all urban scenes and meet the needs of de-precision map NOA. The global vision fusion scheme, which is not lidar, can also be realized by low and medium computing platforms such as Xavier and OrinN.
Picture: blind area image
At the same time, IM AD algorithm has the ability of faster technological evolution. The public test of "go to Gao Jing Map NOA", which will be launched in September this year, will gradually realize the switching of map technology schemes and achieve a non-breakpoint auxiliary driving experience without relying on high-precision maps.
In the past three years, Zhiji has rewritten almost all the underlying code of the smart architecture and Momenta intelligent algorithms, according to Liu Tao, joint CEO of Zhiji Automobile. This is the core competitiveness, but also let Zhiji car master the ability to follow by example and quickly iterate. They allow Zhiji to grow exponentially in the speed of intelligent iterations in the future.
Moreover, in terms of safety, according to the actual driving statistics, the safety of Zhiji IM AD has reached 3.2 times that of human driving. Based on Chinese roads, the IM AD of collision accidents is only 0. 6 times per million km, which is much lower than that of human driving. IM AD's safety class error braking frequency is less than one million kilometers, which is 5 times better than the industry head standard. Most traffic accidents are caused by novice drivers who have been driving for less than a year. NOA, the city of Zhiji, can avoid these situations while being safer than human driving. In addition to smart driving, Zhiji's more than 30 intelligent innovative functions have to be rigorously verified to meet the highest safety standards before they can be launched.
Figure: light effect guidance at navigation intersection
According to the plan, Zhiji will open the commuting mode in Baicheng in 2024 and realize the "Door to Door" full-scene commuting in 2025, stepping into the "urban NOA era". There is reason to believe that after entering the first echelon in just two years, Zhiji, who is better than Huawei and XPeng Inc., can bring to everyone the autopilot with a sense of intelligence, a sense of the future and a deep integration into urban commuting as shown in the initial video.
