Chris, first of all, thank you for sticking with the conversation even though I am somewhat vigorously defending a differing opinion.
I agree that in a general sense, navigation over large areas that are not already covered by a correction system, is not so suitable for RTK GPS. That is not to say that there are not many situations where RTK is IS suitable.
Other than the speed at which you might leave the effective influence of a base station, I am not aware of any limitations of RTK per se, that apply to a navigation. As with all GPS, there are situations where it just isn’t going to work and we have to manage these regardless.
Let me tell you about my rover application since you have gone out of you way to make it sound quite cumbersome and ineffective. My ‘base station’ does not require any setup at all as it is part of a continuously operating network that is freely available as I have already posted. This base station is set up on a first order survey mark and as such, is about as accurate as you can get. I have used this base station at up to 30km away with better than 10cm accuracy with a RTK fix solution. I can plonk my rover down any where within approximately 3000 square kilometres and achieve this with no prior setup. Communication is via a cellular data. Right at this point, within minutes of arriving, I can record way points using the rover itself and have it reliably repeat the intended path. Using this same RTK system I can survey the operational environment to establish navigation boundaries and automate complete coverage of pasture within a field. Generally my rover knows where it is to better than 10cm with a fix solution and better than 1m with a float solution. With standard GPS this entire system just was not reliable enough. I would have to leave wide margins around fences, would have large holes in my coverage and could not reliably navigate autonomously through gateways between fields. The RTK system I use costs less than $100 more than standard GPS.
OK, you’re using the network solution. You’re not using a standalone base station with a VHF/UHF radio link. And you have a surveyed spot for your base. As I have said this has been proven to be more accurate and much faster. It works in my airplanes at 140kts application speed with about 1.5 second latency on actual position vs GPS position.
You are also not using the setup being commonly sold to drone enthusiasts with a standalone straight-line base station with a radio link.
I’m sure you’re aware of all the hassles required with setting one of those up properly.
Chris, at least now we can agree on what a useful RTK setup can look like.
With surveying we setup base stations on regular (if not daily) basis. For the average person who is looking to setup a precisely located base station, it would be all you describe and more. Surveying carries a very large liability risk and as such, there are a multitude of checks and balances in place to verify accuracy. A surveyor will never establish a point in isolation but rather relies on an established network of permanent control marks typically buried in the ground. The predefined relationship between these marks allows us to add additional marks while tracking and minimising error.
Having said that, I am not so sure that a precisely located base station is required to be useful. In a survey job that requires a high degree of accuracy, a base station need not have accurate coordinates. We will often simply place the base station at a convenient, accessible and visible location and have it accept its uncorrected location. As part of the survey we locate and take GPS readings on existing known control marks and this defines the error which can then bee applied to new points of interest to get an accurate location. This is possible due to the RTK solution always being relative to the base station.
When I was initially testing RTK using the M8N and Rtklib, I mistakenly manually entered the base station location using minutes/seconds rather than decimal minutes. The accuracy and repeatability of the solution was fine however it was offset with respect to local survey marks I was testing against. When I switched to using the reported location in the RTCM data, it fixed the offset and I this is when I noticed the entry error. The point being, as long as you have a single fixed reference mark where your base station is set up, you can cover a reasonable area accurately and repeatedly. If it is sufficient to undertake your project solely in these terms and accurate global coordinates are of no particular interest then this can certainly be useful.
As for “not using the setup being commonly sold to drone enthusiasts”, I’m pretty sure that same system I am using is readily accessible using this gear as well. As far as I can see, all these systems accept RTCM corrections via a serial port of some sort. Anyone with a cellphone running an NTRIP client, OTG cable and an FTDI adaptor can have a standalone system using internet based corrections from a local station. Anyone running a GCS with an internet connection can do the same. I forgo the phone and use an ESP32 and a GSM module.
So instead of telling people why it doesn’t work, why don’t we tell then what does work?
Aerial application also carries huge liability risk with the EPA for not being on-target or even applying the wrong rate of a product.
I think we have a couple issues:
the average drone consumer knows they have GPS. And most think GPS is 100% reliable (it’s not)
we have somewhat deceptive marketing going on from manufacturers selling these consumer-grade RTK systems.
So what happens is somebody buys one of these things and sees the 1cm accuracy advertised. Oh wow - it’s way better than regular GPS! So they now expect that their drone is going to hover accurate to 1cm and they’ll be able to navigate around geographic features accurate to 1cm. Except that’s not the way it works.
RTK was not designed to navigate your autonomous vehicle within 10cm of a geographic feature because it can’t. As soon as you have movement involved you have measurement error. RTK was designed to provide repeatability (we call it “traceability”) for surveys and pass-to-pass accuracy for precision agriculture. The instantaneous DGPS solution is repeatable at any one point in time, assuming clear signals. But it drifts. If you make a ground track or pass one day, come back the next, and use the same GPS solution, it may be off by 2 meters. RTK will get that down to sub-meter.
Is the average consumer buying these things going to realize this? No. They don’t know how it works and all they see is the marketing claim.
It is only going to achieve the desired results by a limited set of professionals that know how to set up and use it. And those pros will likely select a more high-end UAV solution from Trimble or Honeywell Aerospace if their job depends on it.
And this is why I said for the people these thing are being marketed to, regular GPS is just as good. That is in no way shape or form trying take away from the capabilities of RTK correction systems. It is a simple observation of the people they are being marketed to not understanding the details of how to use it. So when we have a user that bolts one to a helicopter and hovers it and observes “it doesn’t really seem to be any more accurate than M8N”, of course not. That’s not what it was designed for. So where did that expectation come from? See problem #2 above.
I am a newbie, but I can confidently say that for a practical application of ArduROVER where accuracy of at least 30 cm or so is needed, RTK seems to be essential. I have a zero-turn lawn mower that I am using to mow large fields. I still need to perfect the tuning to some degree but I know that if I lose RTK fixed, the mower misses streaks. With RTK fixed, it mows very precisely (except for my tuning issues). Just my two cents.
Now here is an excellent example of the same thing as precision ag application. Pass to pass accuracy. Since you have a static setup where the base can get an accurate surveyed static GPS solution, it works. If you’re getting more floats than you like, try getting your base antenna higher to keep a cleaner signal and reduce multi-path errors close to the ground.
Multi-pathing affects RTK corrections just as much as it does DGPS.
That’s a VERY cool setup, BTW. I might have to build one of those.
Hi Kenny, very interesting and well done! There are many similarities in our applications. What GPS unit are you using? Where are the corrections sourced from? How are they sent to your rover? Do you use a companion computer?
Thanks! I still have a lot of improvements, especially just making the installation more robust, but it is so much fun to see it run (and useful!) that I can’t seem to stop to make it better! I guess this winter, maybe I will find the time to do so. I have nearly 100% RTK Fixed with my setup. I have my base antenna about 15 feet off the ground with no close objects.
And I agree, that the key for RTK being useful and not much trouble is really when you can setup the base once and your done. Although, I find it not much trouble at all to do another survey-in if I need to move it. But actually, I am really not concerned about absolute latitude-longitude accuracy. What is more important is that it is relatively close to Google Earth images (in Mission Planner). So what I have done is installed the base at the corner of an object I can identify accurately on Google Earth. Then I keyed in that position in the base.
I hope to do more videos soon explaining the changes I have made since my first couple of overview videos. But… having too much fun right now.
Have you ever left your base set for 10 hours or so and see if Google Earth’s georeferencing was actually good? At 10 hours the static GPS solution should be down to less than 20cm. I’ve found instances where Google’s georeferencing has been off by 30 meters. They use USDA imagery in the US quite a bit, which I shoot a fair portion of. But sometimes when the imagery is processed obliquing of the camera, or the GPS unit on the camera (the camera has it’s own RTK GPS) going to RTK float throws it off.
Chris, if that is truly your position, then wouldn’t you have problems advocating the use of Ardupilot generally? The problem of marketing hype is universal. From my point of view, education is the best cure…
If one thing is for certain - you can not rely on the accuracy of Google Maps on the scale we are talking about! Even when some points on a given image are close, distortion due to perspective and vertical elevation can be significant. Just look at some tall building and how the roof outline can be significantly different from its footprint on the ground. Like Chris says, sometimes you can be tens of meters off… @ktrussell, is there any reason you could not determine the limits of your area of interest by using the “save way point” function and by driving the mover around the perimeter?
Chris, my base stays on all the time. It has been on for many days. I really have only begun using it for serious mowing in the last week or so. Before that I was just tuning and working out bugs in the middle of a field. I can’t say how accurate or stable the Google Earth images are for me from day to day. I am being very cautious and only letting the mower work in large open areas at the moment.
And Jim, I absolutely can and have used the save way point feature. I expect that will be the way I will do things when I get a routine established.
Why? ArduPilot is an excellent autopilot. And I do have problems advocating use of certain features of ArduPilot, which I believe depend too much on automation in critical flight stages for helicopters. Especially from a safety standpoint with the larger more powerful piston machines I fly. And I’m quite vocal about why I think that. The development push is to have the system be fully autonomous, which is fine. But it breeds complacency with a system that must be supervised with manual intervention required when it screws up. And one of the major areas where it screws up is related to the weak link - GPS. The system depends too heavily on it. GPS is not 100% reliable and when a system depends on it the system is therefor not 100% reliable. It requires a human operator. It’s not to Terminator III-Rise of the Machines quality yet.
Why? Because RTK is an excellent mechanism for improving GPS accuracy but you are concerned that people are ignorant, have unrealistic expectations and that there could be unintended consequences. The same rationale extends to Ardupilot itself…
Let’s not say “ignorant” as that seems rather harsh. Let’s say not aware of the limitations.
I tend to think along the lines of aircraft applications because that’s what I do. With ArduPilot and helicopters I strongly advise against using GPS augmented flight modes close to the ground because it is not reliable. Can RTK being used at a takeoff/landing zone for autonomous helicopters improve the reliability like the FAA once proposed for precision approaches (and eventually abandoned because it doesn’t work reliably)? Now, that I don’t know because I’ve never tried it. It seems to me that there’s other things could be done in the software to improve that, like limiting the satellite constellation to a certain number of degrees above the horizon below a certain altitude. Or totally disconnecting the GPS solution altogether close to the ground and just use inertial nav for the autopilot. Or use other types of sensors like optical, sonar, radar.
I might have to buy one of these CUAV systems and experiment with it to see if it can do a precision approach around obstacles for a helicopter, and do it reliably. In full-size it has been proven not to work with much more advanced equipment. But let’s say a person operates a fleet of autonomous helicopters that do, say, some sort of delivery service in emergency situations or something. Can a prepared staging area with a takeoff/landing zone be done where the helicopters have to fly a prescribed precision approach to landing for traffic seperation? And use the local RTK correction solution to make the approach and landing reliable? The helicopters would be flying at slow speed on the approach. And transitioning from a safe altitude to critical altitude where DGPS is no longer reliable. That would be a fun experiment. I can think of a couple practical applications for that.
A great discussion, thanks guys for the detail you have put into the RTK question.
I would like to add one piece of information for others reading this blog.
The discussion is about SINGLE FREQUENCY GNSS systems, L1 to be precise.
This is the common consumer grade GNSS gear that is so cheap these days.
RTK can certainly help these systems and get you down to the 5cm to 10cm range as @jimovonz has so thoughtfully shown in the great work he did with his example.
But single frequency GNSS cannot resolve the ambiguities involved with satellite location, no matter how good the RTK or location service fix is.
For Sub cm GNSS you will still need a multi frequency GNSS system to resolve those ambiguities.
I just wanted to add this to clarify the context of the discussion as being about single frequency GNSS only.
Thanks for taking the time to air this subject more.
Gentlemen, I am a fixed wing operator, large combustion engines and I am very pleased to have this type of information made available. I thank all of you for your time and effort here. As Chris had pointed out… I was about ready to get a GNSS with RTK for my Ground Control Station in anticipation of those 1 cm accurate spot landings. Thanks for saving the time, money and disappointment. This being said… I think a 20 to 30 cm radius landing would still beat the 3 meter alternative. If I can set up an RTK linked to a cell phone and get a fix in 2 minutes as Jim Ovonz has done in his sample comparison then that is a worthwhile move in my opinion. as Chris said, it’s network based. I have to admit I dont have an in depth understanding of this yet, my question remains at to the what are the network provided fixes one must pay for and the potential of RTK being ineffective due to the fact the aircraft is moving and if speed is a factor as the aircraft flies at 16 to 22 meters per second. Anyway… good information. Kenny Trussell, awesome gear there… please keep sharing those videos as I need one of those out here in Thailand and yours is inspirational. Thx gents. Craig.
