3D Obstacles "forgotten" if there is a gap between 3d camera FoV and the robot base

Question

What's an algorithm or technique that doesn't forget about obstacles in the following situation?

I've observed obstacles mistakenly cleared using the VoxelLayer costmap_2d plugin on ROS Jade. This happens when there is a blindspot on the floor between what the 3d camera sees and the robot due to the camera's position.

Here's a side view of a ToF camera with a 90 degree field of view. The red dots are points in the point cloud sensed by the camera. The doted lines are rays from the points to the camera.

Here's a view of how VoxelLayer marks the voxels. Voxels can be free (green), occupied (yellow), or unknown (white). Voxels are cleared (set to free) when a ray passes through them. Voxels are marked (set to occupied) when a point is inside them only if the point is higher than the parameter min_obstacle_height.

Now imagine the depth camera has moved left about two cells. The rays for the points on the ground pass through the previously marked voxels and clear them. The robot has effectively forgotten about the box in front of it.

The robot then attempts to crush the box. Ideas?

---

Originally posted by sloretz on ROS Answers with karma: 3061 on 2016-04-04

Post score: 4

Answer 1

An approach we've used in the past for this is to limit the clearing rays to be not the full width of the sensor scan, but only a subset to avoid clearing partial voxels. It's more conservative, but will require better coverage with your sensors.

That's why you can configure different input topics for marking and clearing separately. You can filter the raw scan to remove the edges and label it for clearing only. And use the raw scan for marking.

---

Originally posted by tfoote with karma: 58457 on 2019-04-07

This answer was ACCEPTED on the original site

Post score: 2