Does Hua Weihai think of automatic road-sense to be in what specific technology advantage should have to soup respect?
Hua Weihai thinks of automatic road-sense to be in the advantage such as the traffic control that should have the algorithm with much shirt-sleeve perception, advanced sensor and model, essence to allow to soup respect, specific as follows:
- Much sensor confluence is felt: Use ” 5 feeling are shirt-sleeve ” plan of entire spectrum sensor, shirt-sleeve and high-definition photograph wait for 27 sensor like radar of head, millimeter wave, lidar, ultrasonic, compose builds radius model of 500 meters 3D real time environment. Wet is photographed when sufferring interference like the head, millimeter wave radar still can penetrate pluvial shade provides target distance and speed information, lidar also can obtain high accuracy to nod the cloud through drip-drop on certain level, in what the soup such as rainstorm falls fraise identifying is led still be as high as 98% , “Eye shot ” dependability promotes 3 times above than plan of simple visual sense.
- Carry Duan Damo framework: Apply ” engine of world of high in the clouds + model of action of Che Duanshi group ” framework, train through huge extreme setting, let AI have the own and decision-making ability in abominable setting waits in rainstorm. The photograph is more traditional and statified type plan, end will be felt to upright model – program – control be in harmony is an organic whole, reduce intermediate information loss, delay when whole reduce 50% , response is more quick.
- Macrocosm brings full control: Drive digital batholith and intelligence deepness is shirt-sleeve, apply the brake of get through car, change direction, the executive layer such as motivation, suspension, implementation optimizes control to the overall situation of car attitude. Face the wet slippery road surface that rainstorm brings about, the system adds road surface newly to get used to AEB function oneself, can feel road surface in real time adherent coefficient is dynamic adjust strategy of apply the brake, can shorten the urgent Buddhist templeput on the brakes of 15%-25% stays a distance, the collision risk that because seeper skids,avoids effectively to bring about.
- Powerful lidar technology: Lidar introduces 192 lines 1550nm wavelengh laser cooperates trends power adjustment is algorithmic, density of rainstorm midpoint cloud promotes instead 40% , exploration distance still can maintain 250 meters in rainstorm, than program of industry main trend many 80 rice early-warning are apart from.
- The technology of millimeter wave radar with accurate essence: Radar of 4D millimeter wave can pass contrail of analytic drip-drop motion, the fraise of the drip-drop of divisional diameter 0.1mm and 15cm, in Guangzhou rainstorm test, to the black of cross driveway polybag identifying accuracy rate is amounted to 98.3% , taller than plan of simple visual sense 37% . China the Pro of radar of 4D millimeter wave that carries 6 to grind oneself for ADS 4.0, farthermost exploration distance achieves 280 meters of above, distance precision promotion reachs 5 centimeters, azimuthal resolution is less than 1 degree, perpendicular angle of visual field amounts to 60 ° , can effective identifying be restricted the stereo spatial information such as tall lever, reduce by accident risk of Buddhist templeput on the brakes.
- Efficient algorithm is optimized: Pass the millimeter class calibration of spatio-temporal synchronism, implementation lidar and photograph data resembling a head 0.01 millisecond progression is occupied right neat, can accurate judgement splatters the athletic contrail difference of spray and true fraise. In the meantime, use cross technology of mode knowledge distill, 1 million groups of data that sunshine tags, through quanta computation is quickened migratory to model of pluvial mist setting, make the system is reduced in the false triggering rate in true soup to 3.7% . In addition, still can use dynamic weight to allocate strategy according to different weather situation, if hail weather reduces lidar weight automatically, promotion vision algorithm is decision-making occupy than, let car hold sober judgement.
- GOD network identifies: Of original creation flexible object semantics library, the metabolic in can identifying rain the special rule fraise such as the paper box, account tarpaulin that is blown by wind. The engineer trains AI model with data of weather of 300 thousand groups of extremes, make its master setting of mist of 127 kinds of rain answer strategy, but essence of life identifies inclined the pedestrian that wears a driveway, electric car and the low and dormant fraise in shade giving rain to wait definitely.