Questions about SLAM

Ive got some more questions about SLAM. I’m writing a report on how we can use SLAM and what the R2As capabilities and weaknesses are with this process. We’re very interested in establishing SLAM into our company’s capabilities
1.What are the uses for SLAM deliverables? Is it accurate enough for design capabilities or is the R2A
primarily for concept modeling?
2.Does SLAM data just supplement aerial LiDAR data?
3.We’ve got a client that is interested in redesigning an old ware house. What should we tell him to
expect accuracy wise?
4. Why is the SLAM processing so expensive? Is this because the algorithm is more complex than
processing aerial LiDAR? Is this done manually or with AI software?
5. I see the SLAM file type is .laz and the trajectory is .txt, similar to aerial LiDAR. What other files can
be generated with SLAM files? Can surfaces be generated with SLAM data
6. Does SLAM data get classified? Like walls, floors, stairs, piping, electrical conduit?
7. Is SLAM data something that is brought into CAD to create line work to map out the buildings
plumbing?
8. Is SLAM data harder to work with in regards to how big the file size is or is it similar to the size of
aerial data?
9. I noticed the run time is limited to 30 minutes for SLAM. Can you merge SLAM data sets to each
other? Is this done by simply overlapping similar GCPs?
10.The SLAM workflow is laid out really well in the Rock articles but what are some of the processing
failures people have been running into? Too close to bare objects? Not enough loop closures? No
GCPs?
11. If we use other types of terrestrial scanners can we process that data on Rock Cloud?
12. Do you have any other sources outside of Rock that I could look at to learn more about the uses
and capabilities of running SLAM LiDAR?
Thanks for any constructive input.

First and foremost, some links:

  1. LiDAR data capture typically requires satellite (GPS) connection while collecting data. SLAM processing allows for data collection and processing without this connection. This means that collecting data indoors, under bridges, under trees, etc… is now possible.

Is it accurate enough for design capabilities. It depends. Some scans will be, some will not.

SLAM processing works by recognizing and matching objects of a scan across time. If the scene you are capturing contains lots of angles/objects within the field of view, then the SLAM processing will do a good job of stitching the scene together. The other extreme is pointing the LiDAR at a flat, feature-less wall. This will confuse the SLAM processing and scan drift will inevitably happen.

  1. Not necessarily. We have customers capturing forestry data where they are only walking a forest, or walking a parking lot, etc…

  2. I don’t have a good answer for that. It really depends on how the data is captured.

  3. SLAM processing is resource intensive and it is using much more sophisticated algorithms. It is not manual.

  4. Once the raw collected data goes through SLAM processing you will have a pointcloud just like any other data capture. You can then order any other deliverables or do any other post-processing you want just the same as aerial LiDAR data. For instance, we have a customer collecting SLAM data around the outside of houses/buildings and ordering Rock Surveyor in order to design and build retaining walls.

  5. We do not currently have a deliverable that will do this. Depending on customer feedback this could be in the works.

  6. Yes, it could.

  7. Generally it is lesser points than aerial simply because you are capturing less amount of time than aerial.

  8. Yes, and yes. For instance, you can capture floors of a building. One floor per scan. Then stack them on top of each other and align.

  9. The biggest issue with the data collection is getting too close to bare objects (walls generally). This happens much less with the R360 because of the 360 data collection. The closer you get to one wall the further away you are on the other side of the sensor.

  10. Yes. You can bring terrestrial data into the rock cloud, merge it with SLAM and aerial or just merge it to multiple terrestrial datasets.

  11. I will have to look into this.

2 Likes

Awesome! Thank you so much for your detailed feedback. Will definitely pass this information on and I will update later if we get some SLAM data. Thanks again!

1 Like

FWIW, we’ve had very little success with SLAM. One working job used 2 data sets that did not align sufficiently (by several feet across the interior of a recycling facility). Upon uploading we got a message that there was “drift” in the data from one USB drive and data corruption from a 2nd drive. We were forced to use a laser scanner to get the project done. Despite two attempts by Rock Robotic support to try to re-align the data, it does not align and the data is not usable for engineering purposes in our case.

Given the startup cost for SLAM we are, needless to say, unsatisfied with the quality and data integrity as our two data sets are obviously at different scales for no discernible reason. My advice would be to use a 2nd data capture method (especially considering there’s no way to verify SLAM data on-site).

2 Likes

Thank you for the insight. Which scanner and software did you end up using?

We borrowed a Leica RTC360 from our sister company (who are doing the engineering). We’re in the market for a scanner to purchase since we have a ton of work piled up that’s need to be done.

The R2A, being a directional scanner, will have the propensity to drift if you get into close quarters. Meaning, if you limit its field of view or the objects it is scanning, it will not be able to accurately detect its movement accurately.

The R360 is much better at scanning in close in environments due to its 360 scan pattern.