I’m working with IRIS-GCS motion planning for collision avoidance and trajectory optimization in a complex environment (specifically the vascular). The environment models are highly complex and non-convex. I found the original method for IRIS is to create collision geometry for obstacles in separate sdf file, such as the half_wall_vertical.sdf

<sdf version="1.5">
  <model name="Robot">
    <link name="Robot">
      <visual name="visual_Wall.019">
        <pose>0.0 0.0 0.0 0.0 0.0 -1.57</pose>
        <geometry>
          <mesh>
            <uri>half_wall_vertical.obj</uri>
          </mesh>
        </geometry>
      </visual>
      <collision name="collision_Wall.019">
        <pose>0.0 0.0 0.75 0.0 0.0 0.0</pose>
        <geometry>
          <box>
            <size>0.25 5.0 1.5</size>
          </box>
        </geometry>
      </collision>
    </link>
    <frame name="frame_AboveHalfWall">
      <pose>0.0 0.0 2.25 0.0 0.0 0.0</pose>
    </frame>
  </model>
</sdf>

These collision geometries are created by simple geometry shapes. But a complicated vascular model composed of many simple convex shapes. So I wonder if there is any other method to handle the complicated model collision detection.

I’m think of using point cloud model, but didn’t find functions for that. I found MeshcatPointCloudVisualizer to visualize point clouds and perception.PointCloud but they don’t have the “Proximity role” as far as I know.