costmap-architecture

The inflation_layer must always be last in the plugins list because it reads lethal and inscribed cells from all prior layers to generate the cost gradient. Global costmaps use static_layer first, then obstacle layers, then inflation; local costmaps use obstacle or voxel layers, then range_sensor_layer, then inflation.

This happens when inflation_layer appears before obstacle or voxel layers in the plugins list. Layers write sequentially into the master grid, so any layer after inflation overwrites the gradient. Move inflation_layer to the end of the list.

Combination method 0 (Overwrite) replaces master grid cells, used by static_layer. Method 1 (Maximum) keeps the higher cost, suiting obstacle and voxel layers. Method 2 (MaxWithoutUnknownOverwrite) behaves like Maximum but unknown cells do not overwrite known values.

No, the observation_sources list within a single layer is not order-dependent. All sources are buffered and processed together during the same layer update cycle, unlike the plugins list which is strictly sequential.

update_frequency controls how often layers process sensor data and refresh the grid, affecting responsiveness and CPU load. publish_frequency only controls how often the costmap is published for RViz or external consumers and can be set lower or to 0.0 to save bandwidth.

Value 254 marks lethal obstacle cells that planners treat as impassable. Value 253 is the inscribed cost where the robot center collides. Value 255 means unknown space, while 0 is free and 1-252 form the inflation gradient biasing paths away from obstacles.