2017 Intel China Industry Summit was held in Suzhou from December 1st to December 2nd and Horizon Robot CEO Dr. Yu Kai said in an interview with "AutoR Intelligence Drive" that Intel Enterprise is a very open and ecological Intel investment though But we also hope that the horizon will be fully developed so that we can decide on our choice of processor whether we use NVIDIA or Intel.
Yu Kai is lucky.
At this year's Intel Capital CEO conference in San Francisco, UK successfully won the capital injection from Intel Capital on its Horizon and completed a multi-billion dollar A + round of financing by the end of the year.
This is the first Intel-based company in China to invest in automatic driving solution provider, both get enraged, triggering endless imagination.
Horizon focuses on the 'brain' of end-devices such as autonomous vehicles and smart cameras, giving them the intelligence to move from perception, interaction and understanding to decision making.
Horizon will use this round of financing to speed up R & D of technology products and commerce, create key technologies with artificial intelligence processor architecture and algorithm software system as the core, and launch system-level embedded labor for major application scenarios such as autonomous driving and smart cities Smart solution.
After the acquisition of Mobileye by Intel, Autonomous Driving Alliance already has a series of car companies including BMW, Audi, Mercedes, Ford, Great Wall, Chang'an and Chrysler, as well as Uber, Drip and other travel companies to join.
The addition of the horizon will be an important step in forming Intel's autonomous driving community.
Two years ago, start-ups driven by autopilot came to capital already favored by startups such as Tucson and Yu Technology.
The time of the autopilot market, but also because of Baidu, Tencent, NVIDIA, Intel and other technology giant's presence is filled with smoke.
The horizon has been established for two years and four months and now has R & D facilities in Nanjing and Shanghai. The entire company now has more than 300 very young employees, most of whom are engineers, covering software, algorithms and hardware.
Thinking of the horizon to build a brain for autopilot programs, including software and hardware systems, Yu Kai that 'just in the car to deploy a brain is not enough, it must have a map, there must be big data, and automatic driving The future must be the biggest scene of the edge calculation, the future of autopilot is the largest cloud computing scene.
Yu Kai is pushing the young company to quickly bring its products to market and commercialize in two ways. One is the ability to move forward and the other is the application scenario.
At the technical level, Yu Kai focuses his core business on the AI processor. As Kay sees it, 'artificial intelligence can compute edges with high performance and make edge computing efficient. And end-to-end learning will become very important.'
And then create a high-performance automatic driving stack of software and hardware has become an important mission Yu Kai.
In the design of software algorithms, the autopilot software framework must be transparent traceability.
What is transparent traceability?
Yu Kai explained that if a self-driving car on the road there is a certain imperfect behavior, or a strange decision with the judge, that decision makers must know what is the reason, if you do not know what is the reason, which will Is a black box system.
He further said: 'Neural networks are gradually being taken seriously, the core of which is that because it breaks down an artificial intelligence system into a white box system, this white box system enables us to utilize and construct a software system that has With all the benefits of deep neural networks, we can control this system and continuously improve the performance of the system. '
Future hardware platforms for autonomous driving need to be fully optimized for software tasks, while deep neural networks based on Intel's FPGAs do more than just compute more fluid and accurate calculations, but they also extract richer information latitudes , Fully coordinated and optimized by software and hardware.
Yu Kai said, 'The PFGA platform, jointly developed with Intel, is an open platform that consumes less than 10 watts of power. It is the first embedded artificial intelligence computer vision processor in China that focuses on vision.'
In Yu Kai's opinion, the number of sensor growth in the future is higher than the population growth rate.
One piece of data shows that a car uses two hours a day and will generate four terabytes of data a day, with a travel time of six to seven hours for a shared trip that will produce 10 terabytes of data.
And a thousand such cars, in fact, the number of data generated per day will be greater than the size of the image search engine.
It can be said that the future autopilot is actually a supercomputer running on four wheels.
In the application scenarios, the horizon will start from China's application scenarios.
Yu Kai said: 'The high-speed scene is of no value to the user. For the moment, it still has to solve the pain of the special traffic situation in China.'
One is parking, one is traffic jam.
Yu Kai seems this is the two more important scenes, he believes that 'many people understand that autopilot is a person in the car's autopilot, and automatic parking is actually a person off the car after the auto-driving, first of all it Very satisfied with our needs, on the other hand, the automatic parking application scene is not on the public road, the public roads have the appropriate policies and regulations, and in the parking lot do not need to wait for the provisions of laws and regulations, it is from the realism '
In the just past on November 28, Horizon completed the first public test of automatic parking.
On a Chery eQ1, the horizon was tested with the vehicle's perception, positioning, decision-making and path planning, and with the vehicle's full-body sensor using only a 16-line Velodyne Lidar.
Yu Kai confident, and hope that the commercialization of products in 2019 landing.
This model can be applied to major car prices.
However, before landing, Yu Kai revealed that it has reached an agreement with a car travel operator and will have 1,000 vehicles to operate next year to collect vehicle travel data.
It can be said from the departure from Baidu, Yu Kai has been on the automatic driving commercial landing with accurate grasp.