自主导航系列21-layered论文阅读

自主导航系列21-costmap

2020-9-10

1.costmap论文研究

Costmaps for Context-Sensitive Navigation 这篇论文阐述了layered maps相对于Monolithic的优点，解读了layered costmap的构造和他的生成过程

标准的layered costmap分为4层

1. static map layer:静态层，是master costmap的底层，可以使用预先生成的静态地图，也可以创建空白地图
2. obstacles layer:激光雷达生成
3. voxels layer:使用3维的方式生成障碍层
4. inflation layer:膨胀出non-lethal cost的区域

由这四层汇总成为master costmap，就这样

2.具体实现

准备阅读论文

For the base_local_planner, nothing should be modified, since the planning will not be affected in any way when exploding the maps in layers.

base_local_planner基本不受分层地图的影响。

在common的costmap中，

inflater: 
   10 robot_radius: 0.35 11 
   inflation_radius:0.35
obstacles: 
   observation_sources: pointcloud_sensor
   pointcloud_sensor:
   data_type: PointCloud2 
   topic: camera/depth/points
   min_obstacle_height: 0.2
   max_obstacle_height: 2.0
   marking: true
   clearing: true
   z_voxels:8 
   z_resolution: 0.25
   max_obstacle_height: 2.0

In this example, we create two layers: a inflater layer, that considers a circular robot with 35cm of radius, and, therefore, an inflation radius of 35cm so it doesn’t collide with anything; a obstacles layer, that takes the pointcloud data (if you are using a laser, please change that here) and passes this data to the costmap.

plugins:
    -{name: obstacles, type: "costmap_2d::VoxelLayer"} 
    -{name: inflater, type: "costmap_2d::InflationLayer"}

在global和local里面也应该加上插件

LOCAL COSTMAP: If you’re not using a layered approach, your local_costma p in its whole will be displayed in this topic.

如果不使用分层的办法，那么本地的代价地图将会显示在这个topic中

参考

代码实现