Deep driving – artificial intelligence technology will change the driverless – Sohu Technology

— artificial intelligence technology will change the driving depth of unmanned technology – Sohu ten years ago, Google began to layout a driverless car project, the expensive optical radar and high-definition map technology as its strategic development goals. Today, these two technologies are still an important pillar of Google unmanned project. Based on the data obtained from the optical radar and the camera, we can locate the unmanned vehicle on the map, but this method has not been able to meet the practical requirements. Driverless cars need to have a good perception and decision making ability to drive in a complex environment or in a changing street, which in itself is uncertain. Now, we mainly rely on artificial intelligence technology – depth study to solve this problem. Unlike in the past, we do not use the default algorithm, but let the system through the example of learning, autonomous learning how to make a correct response to an input. It can be said that for most of the cognitive tasks and some low-level control problems, deep learning is the most effective solution. Driverless cars need to be aware of the system to identify moving objects (cars, pedestrians) and fixed objects (Lu Dengzhu, the edge of the road). It can detect dynamic objects in three ways, including cameras, laser scanners and radar. In these three ways, the camera is the cheapest, but in the past, the image into the object is difficult to detect, so this is the lowest rate of use. Through deep learning, we found that the ability to understand and use these images was significantly improved. Even more exciting is that we found that the task of deep learning makes unmanned technology to make further progress. Multi task deep learning refers to the training system and to identify the lane, cars and pedestrians, the training result is better than the three subsystems of independent implementation, this is because in a single network, information can be shared. Unmanned technology is not entirely dependent on the pre-set map, but only the map as one of the data streams, combined with sensors to obtain data to help the system to make decisions. For example, a neural network system through the map information acquired in advance and the position of the crosswalk, pedestrian detection to pass, this ratio only depend on the image data to be much more accurate. In the past, due to the automatic driving car driving is not stable, causing many people to take the automatic driving vehicle will produce the feeling of motion sickness. Now, deep learning can help us alleviate this headache problem, so that the automatic driving car to learn how to drive human beings and skills, will make the passengers feel more natural. Now, deep learning is only just beginning to emerge in the field of technology on unmanned. But just as it has made a breakthrough in image search and speech recognition, deep learning may change the future of unmanned technology. Editor: Cui Chaoyu source: MIT Technology Review MIT Technology Review’s first book "the top" Chinese technology available now is changing the world scientific fact, most worthy of investor expectations of technology recruitment editing, visual design and Practice相关的主题文章: