Self-service Training Document for Partners

Table of Contents

For Mappers

Accessing RapiD projects in HOT Tasking Manager

HOT has set up a special instance of the tasking manager to test out AI assisted features. Visit the link below to access it. 
https://tasks-assisted.hotosm.org/

Find a project marked with RapiD in the name to start mapping. 

Once inside the project select the Map tab from the project page and use either option to select a task:


Click Start Mapping:


Select RapiD from the editor dropdown and click Start Editor:

Editing in RapiD

RapiD is Facebook's fork of the iD editor with a set of enhanced features including a more complete set of validation checks to help catch issues on AI or manually generated roads. It takes all the power of Facebook's AI assisted mapping and makes it accessible to all levels of mappers.  Upon opening a task area, the mapper will be presented with an additional RapiD Assist layer colored magenta. These magenta ways are road predictions that a mapper can choose to add to the map. When editing from the tasking manager these suggested roads will only load within the bounding box and anything not specifically chosen for inclusion by the mapper will not be uploaded.


To use RapiD, a mapper selects a road then clicks “Use This Feature” or uses the hotkey “A” to bring the predicted way onto the map. Until the feature is brought into the map in this way, it is not editable. From there the feature is like any other newly digitized road and can be altered and tagged. The initial generated data is often in a high quality state but it may need further work to bring it up to a finished quality.

If a mapper decides they don't want to use the predicted road they can mark “Remove This Feature” or hotkey “D” to make it disappear. 

You can also add and remove roads even faster if you hold down “A” or “D” and then click on the magenta road.

To aid in tagging these newly created roads correctly, we have added the hotkey “Shift + C” to cycle through the most common road tags.

With the hotkey “Shift + R” the RapiD Assist layer can be turned off and “U” will turn off normal OSM data. This will allow a mapper to see the background imagery in order to validate the data.


Additionally, we developed a style toggle to aid in the editing process. When the hotkey “G” is pressed, any modified part of a pre-existing road will turn green. This will also highlight any newly added roads (through either manual editing or the magenta layer) as green.

Validation Checks

In addition to the normal validation checks now present in iD, we have added a few additional checks that are particularly useful when working with ML data.

Short road checks

Short road checks look for roads that may need extension or deletion.

Disconnected Roads

We have enhanced the disconnected road check to check when a newly added cluster of roads is disconnected from the already existing road network whereas previously it just looked for single disconnected roads.

Extra nodes near connection

This check looks for nodes near intersections that can possibly be deleted to create a more accurate intersection shape.

Fixes Requiring Human Judgement

RapiD is great at finding and creating roads but it is not perfect. Rather than being complete, the data should be considered as a starting point for human editing. Here are some common mistakes in the AI assisted data to look out for while editing: 

Incorrect junction configuration

Forced cross streets

Two T-junctions that are very close together, can sometimes be digitized by the machine as a forced cross street.  As you can see, these are better represented by two separate junctions. 

Y-shaped junctions

Sometimes the machine may add a node or two on either side of a T-junction. These nodes are superfluous, and can create a very slight kink in the way. Often times, they should just be deleted. These will typically be identified by the validation check “Extra nodes near connection” and can be automatically addressed but some may require human identification. 

Incorrect digitization of roundabouts:

The machine usually roughly digitizes around roundabouts so you could see something like the geometry below when first added.  The mapper will need to add the finishing touches using this workflow.

False connections

False connections caused by alternative imagery misalignment or poorly digitized pre-existing roads:

We use the “Facebook's Map With AI - Maxar Imagery” an imagery layer to generate our roads. Pre-existing OSM data may be aligned to other imagery, which can cause false connections or data errors. In the below example, the offset preexisting data caused a poor quality ML road to be predicted. In such cases the ML road should just be removed.
With Facebook's Map With AI - Maxar Imagery
With DigitalGlobe Standard

False connections through physical barriers: 

Sometimes the ML may make a connection through an area that is blocked by physical boundaries such as a wall, fence or even building occasionally.

These are often observed around commercial complexes, temples, schools, hospitals, parks, factories, etc..
Such boundaries can be very hard to see on satellite imageries and the machine sometimes connects these roads by mistake. In the example below the ML mistakes a roof shadow for a narrow road.

False positives

False positives can be caused when the machine mistakingly identifies a feature like a waterway or building as a road. In this example, a canal was incorrectly picked up as a road so the predicted road should just be removed.


Submitting Data

Once the data is ready to be uploaded to OSM, take a last look at the validation panel for any lingering issues. Then zoom out and check the task for completeness and tagging accuracy. When you are ready to proceed press the Save button. 



Then leave a changeset comment describing what sort of roads you have added. 

Hot Keys

A
B
1Function
hot key
2Hide/Show ML Roads
shift + R
3Cycle road types
shift + C
4Add ML feature
A
5Remove ML feature
D
6Highlight edited ways
G
7Turn off OSM data
U

Known Data Issues

We are still hammering out some of the bugs in our road vectorization during this soft launch. There is no need to report these specific issues as we are already aware and working on them. If you find different bugs or data issues please feel free to report them. 

V shaped way


Short duplicate way


Roads duplicating railroads


Y shaped intersections


Suggested Road type Tagging

Automated road type tagging is still a work in progress. We are actively working to improve our predictions for it though.

Other special features on RapiD

Source=digitalglobe

We automatically add source=digitalglobe to all of the RapiD assisted roads and disable the deletion of the tag.

Facebook's Map with AI - Maxar imagery

We have made the imagery used in the creation of this data available as a layer. This will be your best starting point when mapping but, as always, you may need to toggle through multiple imagery sources to create the best data.


For Project Managers

Add RapiD to a Tasking Manager project

After you create a project, you can add RapiD to the list of permitted mappers:


In the Settings menu you can add RapiD for the mapping stage and the validation stage independently:

RapiD will now appear as an option in the editor drop down list.


Additional Resources