Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizens for the delivery of the clues of Frost Wind Shadow! CTOnews.com news on July 27, CTOnews.com learned from DJI car official account that today DJI car officially announced its new generation of smart driving solution named "ch é ng x í ng". DJI car also announced that the first production car equipped with the basic version of "Chengdu platform" will be officially launched in a few weeks.

"Cheng Xing platform" was first unveiled in the 100 people Club of electric vehicles in early April. the biggest feature of the platform is that with the computing power as low as 32TOPS, the pure visual configuration of 7V / 9V, and through "strong visual online real-time perception, no high precision map dependence, no lidar dependence", the L2 + intelligent driving functions including urban memory driving (32TOPS) / urban pilot driving (80TOPS) are realized. At the same time, the pure visual intelligent driving system also supports the expansion of millimeter wave radar, ultrasonic radar, lidar, high-precision map and other sensors to further enhance the performance and safety redundancy of the system in extreme scenes.

According to reports, in the hardware configuration of the "Chengdu platform", the unique inertial navigation stereo binocular in Dajiang is the most obvious. Through the visual principle similar to that of the human eye, distance information is obtained to generate dense point clouds. The "row platform" can not only identify any type of obstacles in perception, but also achieve better scene generalization in the case of weak dependence on data accumulation. In addition, inertial navigation stereo binocular can also provide data support for other vehicle systems, such as road preview data for adaptive intelligent suspension, environmental information for automatic lights, automatic wipers and so on. Inertial navigation stereo binocular is not only the most obvious external feature of "travel platform", but also one of the most important technical cores of "travel platform".

In addition, the "travel platform" realizes the BEV perception of the integration of travel and parking, and adopts OSP (Open Space Planning, open space decision planning technology), which enables the "travel platform" to improve the traffic efficiency of intelligent driving on the premise of safety, whether in open urban roads, closed highways / expressways, and parking environment.

The "travel platform" can complete the construction of real-time 3D scene and road topology based on binocular observation data and BEV perception fusion, combined with visual inertia odometer and birth cost vehicle posture data. Therefore, only with navigation maps, the basic version of "Chengdu platform" can achieve the navigation high-speed function, reduce the dependence on high-precision maps, and can also cover cities without high-precision maps, support memory driving function, and achieve point-to-point navigation function.

BEV (Bird's Eye View) algorithm is a mature perception paradigm, which uses deep learning network to integrate multi-channel camera input, converts the scene around the car into top view, reconstructs the static road elements and road topology in real time, and realizes high-precision omni-directional perception. Through extreme computing optimization, Chengxing platform has for the first time implemented BEV-aware technology to a computing platform as low as 32TOPS.

The "traveling platform" adopts a unified BEV sensing scheme in high-speed, urban, parking and other scenes, which can run adaptively in high-speed, urban, parking and other panoramic scenes, and has a good effect on dynamic traffic participants and static road element detection. This supports the various functions of the "travel platform", such as pilot highway without high-precision map, memory driving, cross-layer memory parking and so on. It is also the industry's first cross-layer memory parking scheme that does not rely on lidar, officials said.

The function of navigation highway without high precision map based on BEV

Driving function without High Precision Map memory based on BEV

According to the pure visual cross-layer memory parking function based on BEV, "Chengdu platform" has the industry's first mass-produced omni-directional monocular depth sensing technology solution, which can be output through high-resolution omni-directional 3D point cloud, and can support arbitrary obstacle detection and obstacle avoidance in the whole parking scene without relying on millimeter wave radar and ultrasonic radar.

Omni-directional point cloud perception supports the parking lot to deal with any obstacles. The scheme covers the whole scene of parking, including the detection of roadsides, thin poles and suspended obstacles, as well as the use in low-light and glare scenes.

The technical highlights of the "travel platform" also include 4D scene flow technology, 4D pure visual self-marking technology, active security technology based on pure vision, and open space decision planning technology.

Through the configuration scheme based on pure vision sensor, the "trip platform" realizes eight types of capabilities: active safety, driving assistance, parking assistance, pilot highway (navigation map only), memory parking, memory driving, cross-layer memory parking and pilot city area (upgraded version of the configuration and expansion of high computing power SoC, can achieve urban open road navigation assistance function that does not rely on high-precision maps). The details are as follows:

Active security: includes a number of active security assistance functions such as FCW, AEB, LDW, LDP, BSD and DOW.

Driving assistance: including basic horizontal and vertical automatic control, including dialing pole lane change, intelligent obstacle avoidance, man-machine co-driving and intersection assistance and other functions.

Parking assistance: including three parking application scenarios: auxiliary parking, intelligent summoning and remote parking.

Pilot Expressway: in the high-speed / urban express road scene, including toll station traffic, intelligent speed regulation, automatic ramp and other functions, support navigation map / high-precision map / route memory three modes.

Memory parking: the system memorizes the user's teaching route and supports indoor and outdoor parking, including designated, regional and exploratory parking modes.

Memory driving: the system memorizes the user's teaching route, including automatic crossing, traffic light identification, lane change, off-ramp and other functions.

Cross-floor memory parking: the system memorizes the user's teaching route and supports upward and downward cross-layer scenarios, covering full-featured parking scenarios from toll poles to parking spaces.

Pilot city: in the urban road scene, including construction road response, overtaking detour, intersection processing and other functions, support navigation map / high-precision map / route memory three modes.

According to reports, the "Chengdu platform" has supported a variety of product delivery models, which can be delivered on a full stack, or can flexibly reuse customers' existing supply chain and software and hardware resources, and accelerate the landing of intelligent driving capacity according to the actual situation. The first mass production car with the basic version of "Chengdu platform" will be officially released to consumers in a few weeks' time.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.