When an autonomous vehicle is cruising along the road, it must constantly be aware of its location. This might seem straightforward, akin to how a phone map displays your current street, but the positioning demands for autonomous vehicles are far more stringent than those for phone-based navigation. An autonomous vehicle needs to know not just which street it's on, but also its precise position within the lane, its orientation, and its distance from roadside infrastructure and other vehicles. Only with precise positioning information can the autonomous driving system make accurate planning and control decisions to execute maneuvers such as lane changes, turns, or pedestrian avoidance.

In the realm of autonomous driving, positioning can be categorized into two concepts based on different reference coordinate systems: global positioning and local positioning. These two types of positioning can work together synergistically, each complementing the other and serving distinct functions.

Global Positioning: Letting the Vehicle Know Its "Position on Earth"

The primary goal of global positioning is to provide an autonomous vehicle with an absolute, time-invariant position coordinate referenced to a standard system on the Earth's surface. This is primarily achieved through Global Navigation Satellite Systems (GNSS), with the most well-known being the U.S.'s GPS, China's Beidou, and the EU's Galileo.

Global positioning outputs coordinate information consisting of longitude, latitude, and altitude, corresponding to the vehicle's position on the Earth's surface. For example, an autonomous vehicle at an intersection in Chaoyang District, Beijing, can have its global coordinates pinpointed to within a dozen centimeters or even better, thanks to the use of high-precision GNSS combined with Real-Time Kinematic (RTK) technology.

The defining characteristic of global positioning is that it does not rely on the vehicle's previous driving history or internal estimates; it essentially reads the position directly from the external world. This means that regardless of the route the vehicle has taken before, as long as the satellite signal is stable, it can immediately determine its current position.

Global positioning provides absolute coordinates. For instance, when an autonomous taxi receives a trip request, it needs to know the absolute positions of the pickup and drop-off points on the map. The navigation module generates a path based on these absolute positions, and the vehicle's subsequent control system follows the predetermined path. Global positioning serves as the foundational positioning for the entire autonomous driving system, letting the vehicle know where it is within the broad framework of the map. This method of positioning is globally consistent and does not require cumulative calculations.

However, global positioning also has limitations. Although satellite signals have broad coverage, they can be interfered with or even completely lost in tunnels, urban canyons with dense high-rise buildings, or areas with heavy tree cover, leading to reduced positioning accuracy or unavailability. Even in open areas, the position output from GNSS alone can have errors ranging from a few meters to over ten meters due to signal noise, which is not precise enough for autonomous driving. To address this issue, autonomous vehicles cannot rely solely on satellites for positioning; instead, they must use global positioning as a foundation and supplement it with other technologies to improve accuracy.

Local Positioning: Letting the Vehicle Know Its "Precise Position in the Local Environment"

When a vehicle is in a specific environment, global positioning only provides a rough coordinate, which is insufficient for high-speed driving or precise maneuvering. This is where local positioning comes into play.

The core of local positioning is to use the vehicle's own sensors (such as LiDAR, cameras, Inertial Measurement Units (IMUs), etc.) to perceive the surrounding environment and match this perceived information with a map or a previously established local model to obtain the vehicle's precise position within the current local area.

Local positioning emphasizes short-distance, high-precision, and continuous positioning. As an autonomous vehicle continues to drive along a road section, it registers its position changes in a local coordinate system by matching visual or point cloud features with a map or its last known position. These techniques are commonly referred to as Visual Odometry, Lidar Odometry, or more broadly, Simultaneous Localization and Mapping (SLAM) technologies in the field of autonomous driving.

For a visually intuitive example, when a vehicle enters a tunnel where satellite signals may be completely unavailable, it must rely on onboard sensors to determine its precise position within the lane. LiDAR scans the environment ahead and around, matching the collected point clouds with an existing high-definition map to calculate the vehicle's position changes. Similarly, camera-based positioning works on a similar principle, identifying visual features such as road signs, curbs, and buildings, combined with short-term estimates from inertial sensors (IMUs), to achieve precise positioning within a local area.

Another significant advantage of local positioning is continuity. Unlike global positioning, which is greatly affected by satellite signals, local positioning can continuously output the vehicle's relative position to a starting point or reference map as long as the sensors and computing system are functioning properly. It is highly sensitive to short-term dynamic changes, making it particularly suitable for precise control of lateral and longitudinal positions during high-speed driving.

However, local positioning is essentially relative positioning, relying on previous states as references. If an autonomous vehicle relies solely on local positioning for an extended period, its position estimates will drift due to cumulative errors. Imagine if an autonomous vehicle continuously estimates its position using wheel encoders and IMUs; all the small errors will accumulate over time, eventually leading to increasingly inaccurate position and orientation estimates. Therefore, local positioning needs to be periodically calibrated against absolute references such as global positioning or known landmarks.

How They Work Together: Fusion is Key

As mentioned above, global positioning and local positioning each have their strengths and weaknesses. Global positioning provides a world-scale position reference, while local positioning emphasizes short-term, high-precision position estimates. To achieve both global navigation and precise control, an autonomous driving system must fuse these two types of positioning.

Fusion typically employs a technique called state estimation or filtering, such as Extended Kalman Filters or optimization methods. Its role is to combine the absolute position provided by global positioning with the relative, fine-grained position provided by local positioning, preserving the non-drifting characteristic of global positioning while also achieving the high precision and continuity of local positioning. Through this fusion, the autonomous driving system can continuously provide high-precision vehicle pose (position and orientation) in complex environments.

There are various forms of implementing this fusion. In some implementations, global positioning provides a rough initial position estimate, which local positioning then refines. When the vehicle enters an area with poor satellite signals, global positioning temporarily interrupts, and local positioning continues to track the vehicle's position changes; when the vehicle regains global positioning, the current local estimate is aligned with the global reference to correct drift. Additionally, local positioning results can be real-time integrated into the global positioning network using graph optimization methods under a higher-level map framework, making the overall positioning more stable.

Application Scenarios and Challenges

From a practical application perspective, the synergy between global positioning and local positioning is indispensable at every stage of autonomous driving. Their respective weights vary in different scenarios such as urban roads, highways, and rural roads.

In open areas or on highways, global positioning performs relatively well, providing a stable position reference continuously, while local positioning can be used to refine the position within the road. In such cases, vehicle navigation and path planning can primarily rely on global positioning, with local positioning serving as an auxiliary means to enhance accuracy. In urban centers or environments like tunnels and underground garages, where global positioning may be severely weakened or even lost, local positioning becomes the primary means of positioning.

Autonomous driving positioning itself also faces numerous challenges. For example, in variable environments, feature matching for local positioning may fail due to changes in lighting, road obstructions, etc.; global positioning may lose effectiveness in signal-blocked areas. This requires the positioning system to not only have good sensor fusion capabilities but also intelligent scene recognition and adaptive mechanisms.

Furthermore, constructing high-definition maps, updating environmental information in real-time, and handling interference from dynamic objects on positioning are all issues that must be addressed in the process of implementing autonomous driving.

Summary

Global positioning and local positioning are the two core components of the positioning system in autonomous driving, providing position estimates for the vehicle at different scales. Global positioning offers world-scale absolute coordinates, letting the vehicle know its position on the large map; local positioning provides short-term, high-precision, continuous local coordinates to support real-time vehicle control. The combination of the two is the foundation that truly enables autonomous vehicles to navigate globally and drive precisely. Building a robust positioning system is one of the core technologies for achieving safe and reliable autonomous driving.

-- END --