Robot Return Path Planning Using Transfer Stations in Grid Maps

The patent segments the large search space into multiple small grid cells, with each cell containing only the minimal necessary map information and obstacle data for that local region. This segmentation dramatically reduces memory consumption compared to storing the entire large search space in memory, while still enabling complete pathfinding by systematically exploring across grid boundaries when obstacles are detected.

The algorithm performs partial A* searches within limited grid cells rather than executing a complete global search. By conducting only the necessary local searches within each grid and using random direction changes when obstacles block the path, the algorithm achieves sufficient pathfinding without the excessive memory requirements of a full A* implementation on large spaces.

The patent merges the simplicity of random direction selection with the structured efficiency of grid-based local A* searches. The algorithm combines random no-proximity direction changes with systematic grid-cell exploration, creating a hybrid approach that maintains low computational complexity while significantly reducing pathfinding time compared to pure random methods.

The algorithm performs preliminary actions by pre-dividing the search space into grid cells and pre-identifying obstacle locations within each cell. This preliminary structuring enables faster subsequent pathfinding operations, as the robot can quickly determine which grid cells contain obstacles and adjust its search direction accordingly, rather than performing complex computations during the actual pathfinding process.