HomeSystem Architecture
Overview

System Architecture

系统架构与数据流

The diagram below shows the complete architecture of the Autoware Universe localization system. Sensor data enters from the top, passes through various localization algorithms, and the arbiter selects the optimal source. Finally, EKF fusion outputs a unified kinematic state.

100%

Initialization

Monitoring & Safety

Post-Processing

Arbitration & Fusion

Twist Estimator

Pose Estimators

Sensing Layer

initialpose3d

ekf_pose feedback

LiDAR
PointCloud2

Camera
Image

IMU
sensor_msgs/Imu

Wheel Odometry
TwistWithCovariance

GNSS
NavSatFix

AR Tags /
Reflective Markers

NDT Scan Matcher
(LiDAR + PCD Map)

YabLoc
(Camera + VectorMap)

Landmark Localizer
(AR Tag / LiDAR Marker)

Gyro Odometer
(IMU + Wheel Speed)

Pose Estimator Arbiter
(Switch Rule)

Pose Covariance
Modifier

EKF Localizer
State: x,y,yaw,yaw_bias,vx,wz

Stop Filter
(Zero velocity)

Twist2Accel
(Derive acceleration)

Localization Error
Monitor

Pose Instability
Detector

Pose Initializer
(GNSS + NDT Align)

/localization/kinematic_state
→ Planning & Control

Autoware Universe Localization System Architecture Overview

Five-layer Data Flow

LayerComponentsResponsibility
Sensor LayerLiDAR, Camera, IMU, Wheel Odom, GNSSProvide raw sensor data
Pose Estimation LayerNDT, YabLoc, AR Tag, LiDAR MarkerIndependently compute absolute pose
Motion Estimation LayerGyro OdometerFuse IMU angular velocity + wheel speed for high-frequency Twist
Arbitration & Fusion LayerArbiter + EKF LocalizerSelect active source → EKF fusion output
Post-processing & MonitoringStop Filter, Twist2Accel, Error MonitorSmooth output + anomaly detection
Previous
Overview & Design Goals
Next
Package Structure