Meituan’s biggest cost comes from its 500,000 riders, Wang said in an interview. According to Meituan’s prospectus, the cost of meituan’s food delivery business increased 238.8 percent from 5.7 billion yuan in 2016 to 19.3 billion yuan in 2017, while the cost of riders rose 94.8 percent from 5.1 billion yuan in 2016 to 18.3 billion yuan in 2017. That’s more than a threefold increase. It has to be said that the rising cost of manpower has accelerated the landing of unmanned delivery trucks.
Recently, Meituan unmanned delivery team conducted a test in Shanghai Songjiang University Town, which makes people think, can unmanned cars really replace people?
First of all, consider the unmanned delivery dining car into the indoor environment to take food, you need to enter the elevator, during which the size of the elevator may not be appropriate, currently Gemdi Group, Chaoyang Joy City agreed to transform the elevator for Meituan free, but other office buildings, residential areas? If you add to the cost of unmanned delivery trucks, isn’t that a loss of purpose?
Let’s look at the outdoor environment, unmanned delivery dining car needs to consider sidewalks, motor vehicle lanes and traffic lights. In such a complex situation, unmanned delivery dining car needs to realize autonomous positioning and map creation, namely SLAM technology (Simultaneous Localization and Mapping).
SLAM is actually a typical va civilian technology, probe planet in unknown environment for scientific research, in the face of complex large-scale environment, to carry on the real-time remote, must create to complete through realizes the simultaneous localization and map navigation task, such as the United States “opportunity”, “spirit” and “curiosity” Mars lander, etc. Without SLAM technology, the robot will not be able to move autonomously and bump around.
SLAM is mainly divided into laser SLAM and VSLAM. Among them, laser SLAM started earlier than VSLAM and is relatively mature in theory, technology and product implementation. In terms of cost, VSLAM uses cameras to collect information, which is much cheaper than lidar. In terms of application scenarios, laser SLAM is mainly applied indoors, which is not feasible for unmanned food trucks, while VSLAM can work in both indoor and outdoor environments.
At present, there are two approaches to realize VSLAM. One is depth camera based on RGBD, such as Kinect. Its biggest feature is that it can directly measure the distance between each pixel in the image and the camera through infrared structured light or time-of-flight principle. However, most RGBD cameras still have many problems, such as narrow measurement range, large noise and small field of vision. They are mainly suitable for indoor SLAM, but not for unmanned food trucks that need to work outdoors.
There is also one based on monocular and binocular cameras, monocular camera SLAM, that is, only one camera can complete SLAM. The advantage of this is that the sensor is simple and extremely low cost, but the biggest problem with monocular vision compared to other visual sensors is that you can’t get exact depth.
Compared with monocular camera, binocular camera has more functions and can obtain information that cannot be accurately identified by monocular camera. The binocular camera uses bionics principle to obtain the synchronized exposure image through the calibrated dual cameras, and then calculates the third dimension depth information of the obtained 2-dimensional image pixels. Deeping in this field, Mimi Intelligent has carried out service and cooperation with more than 200 domestic and foreign enterprise customers. It won the bid for Mimi binocular camera standard edition equipped with six-axis sensor (IMU) and infrared active light detector (IR). By using the complementarity of camera and motion sensor, It can provide visual SLAM research hardware with higher precision, lower cost, simple layout, and face and object recognition.
At present, the combination of binocular vision and IMU is a relatively optimal solution for SLAM. Six axis sensor (IMU) can provide data for study of visual localization algorithm of complementarity and correction, applicable to the visual inertia odometer (VIO) algorithm research, help to improve positioning accuracy, active light detector and infrared (IR) can help to solve indoor white walls and texture of the object recognition problem, improve image source identification precision.
We believe that the unmanned delivery truck with the binocular camera of Xiaomi has infinite application space. So far, I can think of a few things: First, having such an effective point-to-point model, coupled with the attractive nature of the driverless food truck, makes it a good place to promote, for example, afternoon tea based on historical purchases. Second, when users have demands, they can bring the express along with them when they deliver food, which increases the stickiness of users on the platform while improving user experience. Third, the inevitable problem of takeout is oil and vegetables are not enough. Users even choose to let the robot cook food at home at the designated time.
One key direct to the official website of Small binocular camera: http://www.myntai.com