Is there any general problem in using precise positioning (centimeter-accurate GNSS RTK system meant) for autonomous car guidance given I have a predefined path the car should follow? I mean, autonomous cars were the topic #1 at CES 2016 yet no such simple system seems to have been introduced to date... Of course the "path planning" is only a part of the "autonomous package" and other problems need to be solved (collision prevention etc.) but I really wonder whether something simple like RTK guidance could be used.

An RTK system relies on very little amount of live correction data (about 1 kB/s) and mobile networks are really ubiquitous today so I can not really see a technical problem in such solution given there are enough RTK base stations around.


This question is only about using precise positioning to follow a predefined track in obstacle-free environment. I am not asking about other systems that need to be implemented in an autonomous car like collision prevention etc. (such systems may employ LIDAR or a stereo camera). Surely a collision prevention is a must for an autonomous system but I consider a theoretical case only.

An updated question may be: Is precise satellite positioning accurate enough to guide/navigate a full-scale passenger car in an obstacle-free outdoor environment in the speed of about 100 km/h given I have a precise-enough path prerecorded that is to follow?

Some answers below already say yes, this is a solved problem. It would be nice to have the answers elaborated in more detail regarding the existing solutions (accuracy, possible problems etc.). One of the solutions may probably be the open source APM autopilot which work for rovers too (example by Emlid) but that does not seem to use RTK so the accuracy may be rather low.

  • $\begingroup$ GPS navigation has been around for over a decade. Are you talking about using it to keep a car on the road? If so, what's the position update rate on the device you mention? $\endgroup$ – Chuck Jan 10 '16 at 17:22
  • $\begingroup$ @Chuck: Yes I mean using GPS to keep a car on the road. I am mostly aware of the "hobbyist-level" RTK systems only and for instance NS-RAW can do 20 Hz but I think more expensive systems may do even more Hz. I can imagine if a precise enough HD map is available which contains road lanes then one could use such map to guide the car in the center of a lane for example in cases when a vision-based (camera-based) lane departure warning may fail due to bad weather (bad vision conditions) etc. $\endgroup$ – Kozuch Jan 10 '16 at 19:18
  • $\begingroup$ PS. AFAIK GPS should largely be immune from bad weather and should perform nicely in heavy rainfall/snowfall where I think no vision-based system really can deliver a decent performance. $\endgroup$ – Kozuch Jan 10 '16 at 19:21
  • $\begingroup$ I don't think that position sensing, nor improving map data accuracy, are very useful in autonomous driving. What problems would precise position sensing solve for you? $\endgroup$ – hauptmech Jan 10 '16 at 20:36
  • 1
    $\begingroup$ I don't know any papers looking at path precision vs car speed and positions sensing, but you should be able to find it in a journal somewhere. This paper is not behind an academic paywall and will give you a sense of the state of the art 20 years ago (which would be capable of what you want with RTK type positioning) I think he maxed out at 40Km/h but the truck was a beast and the limit was due to vision processing computation. Open source projects are not usually state of the art. $\endgroup$ – hauptmech Jan 11 '16 at 11:11

Precise vehicle location is not a very useful piece of data in autonomous driving so I would not expect to see many products focusing on it.

Path planning and following for cars is a solved problem. If you have a static environment where nothing changes, any good position sensing will let you control a vehicle to follow a path. However, there are very few situations where the environment is static. Cars share the road with children, construction, animals, and very stupid other drivers.

The problem with autonomous driving is not the navigation (getting from point A to point B) part. The problem is how to "understand" a dynamic environment and take the "best" action.

  • $\begingroup$ I edited the original question - you say this is a solved problem. I am interested in more details about current solutions so may I ask you to elaborate in case you have something to add? $\endgroup$ – Kozuch Jan 11 '16 at 10:31
  • $\begingroup$ @Kozuch, it's easier for future users if you don't update the question based on answers. If you have a follow-up you should ask a separate question. $\endgroup$ – Josh Vander Hook Jan 11 '16 at 17:59

Unless you want to put this product at the market (which I doubt), then I wouldn't hope for an RTK-GPS pair to solve your problem.

It could help you, in the extreme case of a dense road network, if a simple GPS receiver got such a bad reading that it would place in in the next block. But it is unlikely that you will perform in such an environment.

Instead, as everybody else mentioned, you definitely need to encode some collision avoidance algorithm, since the uncertainties (even in a static environment) are always great. You cannot rely on your system to run "blind-folded". You will need to give it eyes somehow.

But then, if you do that, you don't really need an RTK-GPS receiver. A simple one would suffice for most proof-of-concept work. However, if you really got the cash, get a cheap pair for about $1000 (something like Piksi) and make your life (only a little bit) easier.

  • $\begingroup$ I edited my question in case you are interested to follow it again. I am exactly asking for that "blind-folded" solution for theoretical case. I am not really sure a conventional GPS with +-5m accuracy could be used for anything reasonable in the autonomous world and I think it is limited to use in consumer products (smartphones, GPS navigations etc.). A 5m or 1cm accuracy is a huge difference. You can not even say if a car is on the road/lane with 5m accuracy. $\endgroup$ – Kozuch Jan 11 '16 at 10:38
  • $\begingroup$ At this point, I think you should specify is this question is about a full scale car or a scaled RC model. This will affect the outcome of the discussion. $\endgroup$ – George ZP Jan 11 '16 at 13:28
  • $\begingroup$ A full-scale passenger car is meant primarily but I am also interested in RC model scale. $\endgroup$ – Kozuch Jan 11 '16 at 13:30

Not the answer you're looking for? Browse other questions tagged or ask your own question.