Global Path Planning

Global Path Planning is the top-level module in an autonomous driving decision-making system, tasked with mapping out an optimal macro-level driving route from origin to destination within a known global environment (typically high-definition maps).

Think of it as the autonomous vehicle's "navigator". Once a destination is set, the global path planning module analyses map data and traffic regulations to determine the optimal roads and junctions to take—much like setting a route on a smartphone navigation app. Its primary mission is to resolve the fundamental question of "how to get there from a macro perspective", laying down the strategic groundwork for subsequent local path planning and precise vehicle control.

💡 Core Technologies and Evolution

Global path planning relies on dedicated algorithms to identify the best route from map data. Here, "optimal" does not strictly mean the shortest distance; it can also prioritise the fastest travel time, lowest energy consumption, or highest safety margin. Below are several classic and mainstream algorithms:

Algorithm Type	Core Working Principle	Key Advantages	Key Limitations	Typical Applications
A Algorithm*	Combines Dijkstra's algorithm (guaranteeing the shortest path) with heuristic search (guiding the search direction by estimating distance to the destination) for superior computational efficiency.	Highly efficient at finding optimal solutions in static environments, making it one of the most widely deployed global path planning algorithms today.	Computational load can still escalate significantly in highly complex or dynamically changing environments.	The classic and go-to benchmark for global path planning, widely utilised in mass-production autonomous driving systems.
Dijkstra's Algorithm	Employs a "breadth-first" strategy to exhaustively explore all possible routes until the destination is reached.	Theoretically guaranteed to find the true global shortest path with dependable accuracy.	Computationally intensive with relatively low efficiency, as it evaluates a large volume of irrelevant areas.	Scenarios where real-time processing is not critical, or as a baseline standard to evaluate other algorithms.
RRT (Rapidly-exploring Random Tree) and its Variants	Progressively builds a search tree through random spatial sampling to discover feasible paths.	Algorithmically straightforward, efficiently handling high-dimensional spaces and intricate geometric constraints.	The generated path is rarely optimal and tends to be somewhat "jagged".	Predominantly used for local path planning, though ongoing research continues to adapt it for global planning applications.

In summary, global path planning is the fundamental prerequisite for an autonomous vehicle to achieve seamless point-to-point autonomous travel. It establishes the master trajectory for the journey, upon which subsequent local path planning and motion control rely to respond to dynamic, real-time road conditions.

Looking at current technological trends, the A* algorithm and its modified variants remain the mainstream choice for global path planning. Meanwhile, sampling-based algorithms like RRT—thanks to their adaptability in high-dimensional spaces—are more commonly deployed for granular tasks such as dynamic local obstacle avoidance.