NetEase Smart has been invited to experience based on Baidu Apollo platform development of the automatic driving car, their L3 level of automatic driving program using machine vision Sensor and depth learning algorithm, a monocular camera time to collect 320 * 320 screen. According to the scene of the engineers, the use of migrating learning ability, in a new closed environment after a short period of time and less data training, the car can be automatically driven.
Autopilot implementation path is not the same, but the same thing. A car composed of an average of 30,000 parts, to leather this complex system of life, the problem is naturally not small. However, the various manufacturers have to explore a road to their own characteristics to achieve automatic driving. At present the field of automatic driving players 'one-third of the world': Tesla as the representative of the car manufacturers, they rely on ADAS gradually grading Suction Control Valve to achieve automatic driving; to Google as the representative of the technology companies, they through the perception system, in-depth learning to achieve automatic Driving; Intel as the representative of the automotive electronics chip suppliers, they are in the unmanned to do a set of computer systems in line with the body. Path is different because the core technology is not the same, some companies are started to identify the image, and some companies are more good at radar technology, many people according to the need to develop their own perception program.
As long as the auto drive trial operation and intend to the recent commercial landing of the company, most of the use of visual perception or visual-based, radar, supplemented by the perception program, the price of radar is too expensive, and now can not produce enough value. Tucson, who started from computer vision, continued to use the lidar for the time being without the use of a laser radar program. Their technology relied on 10 cameras and three millimeter-wave radars, but Tucson said it did not rule out the radar program. It is important to control costs. But as China's old technology companies Baidu, technology is more flexible, it may be because Baidu did not put pressure for a short time, Baidu Apollo to do Android for car. Baidu L4-level track car is based on the Apollo platform, installed with differential GPS and inertial navigation for positioning and completion of the calculation of the state of the car, and with high-precision map to achieve tracking function.
Former Intel China Research Institute, Yu potential technology CEO Wu Gansha said they use the camera, millimeter-wave radar, laser radar fusion program to achieve automatic driving, in addition to ensuring the safety of trial operation, is still the program comparison, inappropriate The perception scheme will be discarded in the future. The self-propelled driving scheme uses two 16-wire laser radars, located on both sides of the vehicle body, and is equipped with three cameras and four millimeter-wave radars. And high German map executive director Wei Dong Fuel Rail Pressure Sensor told Netease smart, high-precision map with high-precision positioning for the realization of automatic driving is the most important, it is the infrastructure, sensors, lasers and play a fusion role, in fact, on the automatic driving Research, many people into the misunderstanding, such as through image recognition, laser radar to locate, so of course can be done, but it is not the best solution is questionable.
As for the Silicon Valley Google automatic driving with a laser radar-based camera supplement solution; Tesla using Mobileye to computer vision with millimeter-wave radar to achieve unmanned program.