Local Path Planning

Local path planning serves as a pivotal link bridging "decision-making" and "vehicle control" within autonomous driving systems, tasked with generating a precise, safe, and feasible driving trajectory based on real-time surrounding conditions under the broader guidance of the global navigation route.

Unlike global path planning, which relies on fully mapped data to determine "which route to take" before setting off, local path planning is all about real-time tactical adaptation on the move. It primarily depends on live environmental data captured by on-board sensors (such as cameras and radar), including the positions of surrounding vehicles, pedestrian movements, and temporary obstacles. It dynamically modulates the vehicle's driving path whilst strictly complying with physical constraints (such as maximum steering lock and acceleration limits) and road traffic regulations.

Several mainstream technical approaches are used to implement local path planning, each with distinct advantages:

• Sampling-based algorithms: Such as enhanced RRT (Rapidly-exploring Random Tree) algorithms, which search for viable paths by randomly sampling within the drivable space. Optimised algorithms deliver substantial gains in computational efficiency and path smoothness, cutting processing times by 64% to 78%.

• Optimisation-based algorithms: Such as those utilising g2o solvers, which formulate path planning as a mathematical optimisation problem to determine the ideal path while meeting key constraints like vehicle kinematics and collision avoidance.

• Decomposition-based planning frameworks: Such as methods operating within the Frenet coordinate frame, which split trajectory generation into two dedicated sub-tasks: path planning and speed planning. By resolving static and dynamic obstacles independently, this framework enhances computing speed and adapts well to complex driving scenarios.

In summary, global path planning governs strategic navigation, whereas local path planning handles tactical driving execution. Only through the seamless synergy of both systems can autonomous vehicles navigate complex and fast-changing road environments safely and efficiently.