mintar
commented
3 years ago I have made an attempt to do just this two years ago. I added two upward-pointing Pico Flexx cameras to the front and back of the robot in order to prevent collisions with the top part of our robot. The images below show one of the cameras.
Then, I made the following changes:
- run a PCL VoxelGrid filter with voxel size = 10 cm to reduce the amount of data coming from the cameras
- publish the point cloud topics and the tf transforms of the camera frames to the MiR over the mir_bridge
- modify the
move_baseconfiguration on the MiR internal PC to add the point cloud topics asobservation_sources.
This kinda worked, but unfortunately not as well as I had hoped. There were two problems with this approach:
- The processing of the additional point cloud topics really slowed down the internal MiR PC.
- You often get into a situation where you have entered a dynamic obstacle (such as a person) into the costmap, then the robot moves away, and the obstacle moves away as well. However, since the cameras are pointing upwards (and not horizontally), you can never remove the obstacle again unless you drive back to the position where you first observed it. So very quickly, the map gets swamped with spurious obstacles. This makes finding a path almost impossible, and probably also contributed to the computational load on the internal MiR PC.
On the other hand, the MiR supports a very similar use case with its additional "top camera", so it must be possible somehow. But since this was not a top priority for me, I did not investigate further and just gave up after some time. I'm not saying that it is impossible.
If you want to "try before you buy", you could try including additional 3D cameras into the Gazebo simulation and move_base config in this repo. This would at least enable you to figure out solutions to problem 2. Note that I wrote the move_base config in this repo just to be able to run it in Gazebo. It's not identical to the move_base config on the actual robot. Navigation on the actual robot works even better than my config.
If you want, I can provide you the code that I wrote back then.
ComplexSysSolutions
ravescovi
The MiR100 only appears to directly support the use of 2 3D cameras, which provides a limited field of view around the robot. I was wondering if it would be possible to add additional 3D cameras to a host PC running this library and publishing this data to a topic on the MiR PC such that it can use it for obstacle avoidance.
We are currently evaluating the MiR100 for our project so I unfortunately can't test this myself. Thank you.