Building Global and Local Path Planners for Autonomous Robots

Autonomous navigation for robots involves planning both long-term routes and short-term movements. A robot must first determine a useful path through its environment and then continuously adapt that path to obstacles, other robots, people, and changes in the surroundings. To accomplish this, navigation systems divide planning into two distinct phases: global and local.

The global planner operates on a larger scale, considering the overall environment. Its primary goal is to find a viable route between the robot's starting point and its destination. Common techniques used by global planners include A* search, Dijkstra's algorithm, graph search, grid-based planning, and sampling-based planning. The output of the global planner is a high-level path that outlines the major waypoints a robot should follow.

On the other hand, the local planner works in close proximity to the robot, responding to real-time observations. It takes into account the robot's current velocity, size, and the positions of nearby objects. The local planner also considers temporary goals, such as avoiding dynamic objects or adjusting the trajectory based on the current direction of motion. When faced with unexpected changes like a person suddenly entering the robot's path, the local planner steps in to slow down, stop, or navigate around the obstacle.

Both global and local planning are essential because a robot may initially follow a valid global path, only to encounter unexpected challenges along the way. For instance, if a robot's global route leads it through a hallway and a person unexpectedly blocks the way, the local planner's job becomes critical in adjusting the robot's movement to safely navigate the obstacle.

To create an effective navigation system, several key components need to be implemented. First, a reliable map of the environment must be available, either as a grid or a costmap that assigns higher costs to undesirable regions. Next, a global planner must be constructed to search through free cells and assign costs to regions that the robot should avoid.

Following this, a local planner needs to generate multiple candidate trajectories for the robot to choose from. These trajectories should be evaluated based on factors such as collision risk, distance to the path, proximity to the goal, smoothness, velocity, and clearance.

In ROS2, the navigation system can be organized modularly with distinct modules for the map, global planner, global plan, local planner, and velocity commands. Communication between these modules occurs through topics such as /map, /global_planner, /global_plan, /local_planner, /scan, /pointcloud, and /cmd_vel.

Modular design is essential for testing and maintaining the navigation system effectively. In a production environment, monitoring various aspects of the system is crucial. These include planning latency, path validity, obstacle updates, controller frequency, recovery behavior, and localization quality. Additionally, the system should include explicit recovery behaviors when no safe local trajectory can be found.

When designing a practical navigation system, start by ensuring a reliable map, selecting an appropriate global planner, creating a local costmap, developing a local trajectory generator, implementing collision checking, setting velocity limits, and designing recovery behaviors. Continuous monitoring is vital to maintain a robust navigation system that can adapt to new challenges while maintaining safety and efficiency.