ROS Resources: Documentation | Support | Discussion Forum | Index | Service Status | ros @ Robotics Stack Exchange
Ask Your Question
4

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

asked 2016-04-04 17:44:36 -0500

sloretz gravatar image

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.

image description

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.

image description

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.

image description

The robot then attempts to crush the box. Ideas?

edit retag flag offensive close merge delete

1 Answer

Sort by ยป oldest newest most voted
2

answered 2019-04-07 23:08:36 -0500

tfoote gravatar image

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.

edit flag offensive delete link more

Question Tools

1 follower

Stats

Asked: 2016-04-04 17:44:36 -0500

Seen: 655 times

Last updated: Apr 07 '19