I recently read about RatSLAM, a visual SLAM implementation that only needs a simple monocular webcam to map and localize, instead of the usual high precision range finders and encoders that most SLAM algorithms require.
It even has some open source implementations, including a ROS implementation, but it's been unmaintained for years and seems to have gone largely unnoticed. I can only imagine this is because of some severe limitations in the algorithm, but I'm not clear from the paper what those limitations are.
The biggest headache with SLAM are the expensive and complicated sensor inputs and hardware it requires. I would think any method that could work with any cheap junky webcam would have a ton of applications, even if there's a slight drop in accuracy. What am I missing? What are there the practical limitations of RatSLAM that the author doesn't mention?