table of Contents
Original Awa Jun Apollo developer community 2018-12-13
On one, we push the Course ① speak for everyone - "autopilot Industry Overview." This week, according to the second lesson Awa Jun Apollo Product Owner David.Zhou taught "Apollo open source modules explain (on)" were the main points of order, more convenient yet reached the scene / do not have time to watch the video of old iron are reading.
Ado, welcome to the developer went into the second phase of the advanced courses.
Unmanned introduction
Unmanned level of integration of safety and accuracy of complex technical requirements are very high, so far, in civilian areas, the technical difficulty of the unmanned vehicle is one of the best.
Do a driverless car is certainly better than doing a phone APP complex many times. Unmanned vehicle which includes the core technology in the end of it, then I will explain you one by one.
We highly touted driverless car is not an ordinary car, because it must be a driverless car remote control car. The so-called by-wire (by-wire), this means that the car is a computer that can be controlled. Between the steering wheel and the motor vehicle is connected to the motor drive wheel, and the wheel to control the motor, so the computer can be controlled by means of controlling the wheel motor.
In order to ensure the safety of unmanned vehicles, when it is on the road, you must do it with the cloud is connected. From time to time with the driverless car does not need to report to the next will be how to deal with the cloud, but the cloud to tell its location and travel planning.
High-precision map
For unmanned, it must have a high-precision driving on the road map, here's a map with high-precision we usually use car navigation map any different?
The so-called navigation map is what we usually use Baidu map, Google Maps, etc., on the navigation map, it will tell you how from point A to point B, and then a couple of choices, is expected to need much time and other information.
The high-precision map relative to the navigation map, the biggest feature is high precision, it is necessary to do the lane line level. In other words, high-precision map not only to know where you are along the way, you also need to know which lane lines in this way, because the only way to accurately tell which unmanned vehicles should travel in the lane, the next should be how turn.
- First, high-precision map to unmanned vehicles can predict a lot of space. No one knows when the car in front of traffic and traffic sign information through high-precision maps, travel planning can be done ahead of time to ensure the stability and economy of driving.
- Second, high-precision map unmanned vehicles can help reduce the amount of calculation. When the unmanned vehicle through the intersection needs, which advance forward light sensing state, when the high precision it can help map to a specific area where the signal light, thereby effectively reducing the amount of computation of the full range of scan identification.
- In addition, high-precision map of all roads and static obstacles around the collection, processing algorithm to reduce the unmanned vehicle static obstacles.
Locate
After the unmanned vehicle with high-precision map, you also need to know what position it on the map, where you need to use positioning technology. When it comes to location, many people may first think of GPS. So what is the principle of GPS positioning is it? In fact, GPS positioning principle is actually a relative positioning.
Each satellite will constantly send electromagnetic waves when the electromagnetic wave signal is received, the received signal according to the time and distance between the calculated speed of light and the satellite. With distance and after four satellites, you can solve an equation to calculate the location. However, due to ionospheric disturbances, reflex like, GPS positioning is erroneous, it can only reach meter level accuracy. For unmanned vehicles, the meter-level positioning accuracy is not safe, in order to solve the problem unmanned vehicle positioning, but also other technologies.
GPS positioning there is a problem, it is beating. GPS positioning is based on the current time all the time to the count, where the results are not allowed to easily occur. In order to smooth the transition of GPS, the need to use IMU (inertial navigation), a general GPS and IMU are used, to keep to a GPS direction to IMU calibration, GPS and IMU give a direction.
In addition, in the case where no vehicle receiving GPS signals, it is necessary to use another technique - geolocation.
GPS positioning and geometric principle of the principle is similar, that is, choose a few feature on the road, where the car unattended calculated based on these feature position. Geometric positioning of high precision, can accurately calculate the position where the unmanned vehicle. So, the more popular positioning technology is the integration of GPS, IMU and a series of geometric positioning technology.
In order to improve the positioning accuracy of GPS, and everyone on the basis of GPS RTK technology it invented. GPS positioning in mobile phones, mobile phone is a mobile station, it will often change position. The RTK mentioned here, is a stationary station, it also receives the satellite signal. The unmanned vehicle RTK and not too far apart, the interference between the two signals by difference smooth, it can be considered unmanned vehicles and RTK signal received is the same.
Application of the RTK positioning accuracy of GPS allows up to 10 cm, but the application of the RTK technique has a limit, it requires from the base station to the vehicle is less than 16 kilometers.
Perceive
We humans have the innate ability to perceive the outside world, because we have all kinds of "sensors."
No one wants security cars on the road, they must also have various types of sensors of its own to perceive the environment. Currently, the main sensor unmanned vehicle with a camera (Camera), a radar (the Radar) and a laser radar (Lidar).
Camera (Camera)
For unmanned vehicles, observe traffic signs and some lights need to identify the color .
The camera just to solve this problem, its main advantage is the ability to distinguish between colors, but it does not have the ability to judge distance . Therefore, the unmanned vehicle perception of the outside world also need to meet other sensors.
Radar (Radar)
Radar is one of the more important unmanned vehicle sensor, which uses electromagnetic waves to detect the target electronic apparatus.
Radar emits electromagnetic waves irradiated to the target and receives an echo thereof, thereby obtaining a target to an electromagnetic wave from the emission point, range rate (radial velocity), azimuth, elevation and other information . Radar application scenarios are all-weather, due to the electromagnetic waves can bypass some of the stuff, so its accuracy is not too high.
Radar also has a feature that determines the speed is very accurate, but more false positives for static objects.
Laser radar (Lidar)
Lidar technology is currently a more central unmanned vehicle development, which is based on the feature amount of the position of emitting a laser beam to detect targets, the radar system speed.
After the laser radar works transmits a probe signal to the target, and then received from the transmitting signal reflected from the target back to the comparing, for proper treatment, the target information can be obtained.
The biggest advantage of laser radar to determine the distance is very accurate, but it also has a big drawback is very high demands on the environment, such as in fog and haze in laser radar accuracy will be reduced a lot.
Lidar is a very new technology, before the development of unmanned vehicles, it does not have large-scale commercial application scenarios, so it's very expensive. Although the laser radar is very expensive, but still with a lot of its kind.
You may have seen the car has no non-stop rotation of Lidar, this is mechanical radar. Because it requires constant rotation, which itself is large and heavy, mechanical loss is large, prone to problems. Based on mechanical radar, we invented a new laser radar, such as MEMS lidar, flash lidar and so on.
After the unmanned vehicle to receive information from various sensors acquired needs to be done sensor fusion . The so-called sensor fusion, it is to see all the things each sensor are added together. When information is integrated with all the sensors see, no vehicle will be able to more fully specifically perceive the external environment.
Trajectory planning
When the vehicle is traveling on the road, by obtaining the sensor information around the next need to know how the reaction.
For around the vehicle or other obstacles, no car is needed to stop or avoid the need, or need to go beyond and so on, these are constraints. No car required in all the constraints, the planning can take out a route.
Unmanned vehicle trajectory planning:
- One is to satisfy all constraints.
- The second is to ensure smooth movement of the vehicle. The so-called smoothing, is to ensure that the car's speed can not jump.
- On this basis, the acceleration of the car also need to be smoothed.
- The ultimate aim is, in human perceptible range, driving the car is smooth, not smooth smooth No situation.
control
As mentioned earlier, no remote control car is a car, not just a collection of gears.
Achieve control of unmanned vehicles, we need to know the relationship between the brakes and slow down, step on the gas and accelerated relationship, when the unmanned vehicle to get some control parameters, you can achieve computer-controlled unmanned vehicles.
Cloud
No one wants to ensure that the car is very safe, we must do so that every one unmanned vehicle to become "experienced driver." The so-called experienced, is seen a lot of traffic and know how to handle these roads.
Baidu cloud in the background there is a huge space simulation, each one unmanned vehicle can upload your own experience of complex road to the cloud, the Internet will have a very large database.
When the unmanned algorithm updates, you can run it in the cloud simulation scenarios, whether the test can meet these road cloud. This step is to ensure that each of the unmanned vehicles are regarded as "an experienced driver."
﹏﹏﹏﹏﹏﹏﹏﹏ END ﹏﹏﹏﹏﹏﹏﹏﹏
