Is OSM good for autonomous vehicles?

Just curious…

Been reading about autonomous vehicles and their need for accurate mapping solutions.
Would OSM maps be considered accurate (of course they may not have enough coverage for many places) but the currently drawn roads and highways, are they accurate enough for autonomous vehicles?

I would say “no”.

In most cases the width of the streets and the lanes are missing, the ways are supposed to be drawn in “the middle of the road”, but this is often not the case. The placement tag is not used everywhere where it could be used, lane transitions (e.g. from 2 to 1) are not properly mapped etc.
There is a lot to map before the map can be used for autonomous vehicles.

I think our data is to inaccurate for autonomous vehicles because our mapping techniques are too inaccurate. Consumer GPS has an accuracy of about ± 10 meters. Aerial imagery has offsets and distortions.

No map by itself is accurate enough for autonomous vehicles. Real time data about the road and lots of cameras/sensors are needed.

also, read the thread I started about GPS accuracy. That speaks a bit towards what OSM contains vs how much it would have to be out of variance to need adjustment.

I wondered so as well. How does one discipline ‘middle of the road’ drawing?
Even if not used for autonomous vehicles (AVs?) maps should be as accurate as possible?

How does Google or Here get it to that level of drawing accuracy? Technology? Their mapping vehicles? Are their maps drawn by humans?

(My Columbus V900 gives me fairly accurate ‘middle of the road’ tracks over unmapped roads when driving. But. In JOSM when I ‘simplify’ the track, it loses accuracy, resulting in me having to draw it almost all over again. Just wondering if Google and Here use these high resolution tracks without simplifying?)

Remember that Google has driven many of those same roads, with very accurate GPS units, coupled with cameras. They have essentially made their own tracks.

Here HD (Nokia’s solution for autonomous cars) uses Lidar (laser technology, 10-20 cm accuracy) to measure the whole environment.

@RayOnTheFarm so Google’s vehicles auto create the tracks as they drive and collect data?
Let me do a search on this this…

In the UK, Google get their maps from Ordnance Survey, who will have used professional presumably DGPS, surveying equipment. Even then road positions are only guaranteed to single figure metres, not enough to get you into the right lane.

Autonomous vehicles will not be using such sophisticated equipment and will not be able to choose their timing to get the best satellite configuration. I therefore have to assume that autonomous vehicles use optical means for lane tracking, like wetware controlled vehicles do.

A long term future might be to have machine friendly guides embedded in the roads.

It’s also worth noting that GPS is jammable. Criminals have done this to defeat satellite trackers and to aid in falsifying lorry movements. They are probably using jammers that will cause a complete loss of signal, but the civilian GPS signal can be jammed in a way that makes it still look valid but distorts the position fixes. (Military system are encrypted, and you need to know the key to produce a signal that looks valid.)

Even I think that is perhaps one of the better ways to solve this.

Wow! Bringing back a five year old topic.

FWIW, sometime fairly recently I read about a MIT group that successfully used a OSM base map and a AI vision system to navigate around some rural roads. The point of the article being that instead of mapping the roads to microscopic accuracy they were able to use the map as a human would: To give them a general idea of the topography. But the actual driving, etc. was done via cameras and AI.

IMHO that is the only acceptable way to use any map. The automomous vehicle always has to consider what it finds in the real world, the map just makes it easier to find out how to get to another part of the world which is not yet visible for any sensors.
Funny thing is that some human drivers show what happens when you trust a map without thinking and without using their sensors :wink: