How much do you know about autonomous driving inertial navigation technology?

Inertial navigation is generally integrated into GPS equipment and is integrated by suppliers. So what is the need to discuss here? We must know that when the vehicle is driving, we can get the yawrate and speed signals of the GPS. Moreover, the vehicle itself has a set of sensors to obtain yawrate and speed, and because trajectory estimation is an important part of autonomous driving, understanding the working principle of inertial navigation can help us do vehicle body-based trajectory estimation.

Inertial Navigation

At present, the integrated navigation system composed of GNSS IMU is the mainstream positioning system solution, and the inertial navigation system is the only one that can output complete The equipment with six degrees of freedom data has high data update frequency and is the fusion center of positioning information.

The core algorithms used in inertial navigation mainly include three types: 1. Inertial navigation solution algorithm; 2. Kalman filter coupling of integrated navigation. 3. Integration of environmental feature information and inertial navigation.

Integrated navigation system core algorithm framework

Hardware and principle

The inertial navigation system (INS) uses the inertial sensor (IMU) to measure the specific force and angular velocity information of the carrier, combined with the given initial conditions, and integrates it with information from systems such as GNSS to perform real-time An autonomous navigation system that estimates speed, position, attitude and other parameters. Specifically, the inertial navigation system is a type of dead reckoning navigation. That is, the position of the next point is deduced from the position of a known point based on the continuously measured heading angle and speed of the carrier, so that the current position of the moving body can be continuously measured.

Inertial system working principle diagram

The inertial navigation system uses an accelerometer and gyro sensors to measure the motion parameters of the carrier. Three vertically arranged gyroscopes are used to measure the angular velocity of the carrier around its three coordinate axes, and are also sensitive to the angular velocity of the earth's rotation.

The accelerometer is based on Newton's second law and uses capacitive, piezoresistive or thermal convection principles to obtain the acceleration value by measuring the corresponding inertial force of the mass block during the acceleration process. Used to measure the acceleration of each axis on the moving body coordinate system.

Inertial system working principle diagram

Inertial navigation through the gyroscope The measured angular velocity is integrated and transformed to calculate the attitude angle (roll, pitch angle) and azimuth angle of the vehicle body. The components of gravity acceleration on each coordinate axis can be calculated based on the attitude angle. The acceleration of each axis measured by the accelerometer is integrated after subtracting the gravity acceleration component to obtain the velocity and position. The state calculated by inertial navigation is used to predict the current position of the vehicle, and then compared with the position (or observation data) obtained by the satellite positioning receiver. The compared deviation includes the inertial navigation estimation error and the satellite receiver positioning error. After weighting through the data fusion algorithm, it is used to correct the inertial navigation prediction, making the inertial navigation prediction more and more accurate.

Inertial navigation solution algorithm

Usually divided into the following steps:

• Attitude update: Integrate the angular velocity output by the gyroscope to obtain the attitude increment, which is superimposed on the last attitude;
• Coordinate conversion: from IMU From the carrier coordinate system to the position and velocity solution coordinate system (inertial coordinate system);
• Speed ​​update: it is necessary to consider the removal of gravity acceleration to obtain the acceleration in the inertial system, and obtain the velocity through integration;
• Position update: get the position through velocity integration.

Principle diagram of inertial navigation solution algorithm

In In inertial navigation, each iteration of the navigation equation needs to use the last navigation result as the initial value, so the initialization of inertial navigation is one of the more important parts. Attitude alignment refers to obtaining the roll, pitch, and yaw of the IMU. The alignment process of roll and pitch is generally called leveling. When the car is stationary, the specific force measured by the accelerometer is only caused by gravity, which can be solved by f=C*g; for a very high-precision IMU, the compass alignment method can be used. When the car is stationary, the specific force measured in the carrier system is The rotation of the earth is used to determine the orientation (yaw) of the carrier.

Inertial navigation initialization schematic

Kalman filtering of combined navigation The coupling of the filter

uses the coupling of the Kalman filter to fuse the IMU and GNSS point cloud positioning results. It can be divided into two methods: loose coupling and tight coupling.

The loose coupling filter uses the difference between the position and velocity measurements and the calculated position and velocity as the input of the combined navigation filter, which is the quantity measurement of the Kalman filter. Tightly coupled data include GNSS navigation parameters, pseudoranges in positioning, distance changes, etc.

Loose coupling schematic diagram of Kalman filter

Tight coupling schematic diagram of Kalman filter

Comparison of the advantages and disadvantages of loose coupling and tight coupling of Kalman filter

Taking the inertial navigation system used by Baidu Apollo as an example, the loose coupling method is adopted. And an error Kalman filter is used. The results of the inertial navigation solution are used for the time update of the Kalman filter, that is, prediction; while the GNSS and point cloud positioning results are used for the measurement update of the Kalman filter. The Kalman filter will output the position, speed, and attitude errors to correct the inertial navigation module, and the errors during the IMU period are used to compensate for the original IMU data.

Loose coupling of Baidu Apollo Kalman filter

Kalman filter fusion diagram

Integration of environmental feature information and inertial navigation

The positioning accuracy and stability of the currently commonly used GNSS IMU combined inertial navigation solution in some scenarios still cannot fully meet the requirements of autonomous driving . For example, in scenarios where GNSS signals are weak for a long time, such as urban building groups and underground garages, relying on GNSS signals to update precise positioning is not stable enough. Therefore, new precise positioning update data sources must be introduced, and lidar/lidar/ It has become an inevitable trend to integrate visual sensing positioning and other environmental information for positioning.

Schematic diagram of an architecture for integrated navigation and environmental awareness information fusion

Take Baidu Apollo's multi-sensor fusion positioning system solution as an example. The inertial navigation system is at the center of the positioning module. The module fuses IMU, GNSS, Lidar and other positioning information, and the final output after solving and correcting the inertial navigation system satisfies High-precision position information with 6 degrees of freedom required for autonomous driving.

Baidu Apollo’s inertial fusion positioning module framework

The above is the detailed content of How much do you know about autonomous driving inertial navigation technology?. For more information, please follow other related articles on the PHP Chinese website!