Request a suggestions in Cost effective solutions in GNSS RTK for autonomous rover

Hi everyone,

We are currently developing an autonomous rover. To achieve precise localization and mapping, we are implementing GNSS RTK technology in stationary base and rover method. We are seeking guidance on what suitable products available and setup procedures to attain centimetre-level accuracy. We requires heading accuracy in a rover board to ensure accurate navigation. Cost-effective products are preferred.

We would greatly appreciate it if you could provide detailed information on products that meet these criteria. Additionally, any instructional materials, such as videos or articles, that offer guidance on setting up a GNSS RTK system would be highly beneficial.

Best regards,
Rakesh.

In what kind of environment will drive the rover? Fully open sky or not?

yes, in a open sky running through open fields

I’d think the first decision would be to decide if you need to implement a Moving Base or not, depending on your heading accuracy requirements.

A single-antenna RTK Rover (while moving) can usually suffice, but that really depends on your navigation goals.

If you plan to use an off-the-shelf autopilot, then your choice in autopilot will also help direct you.

Our goal is to navigate through agricultural fields in autonomous mode using a snake-like zigzag pattern to ensure full field coverage with centimetre-level accuracy. To achieve this, we need to determine the current position for navigation, and for that, we are using RTK. The base station is stable, but we require improved heading accuracy to guide the rover in the desired pattern. Is using an autopilot system the best option, or are there other cost-effective alternatives available? We are looking for suggestions on the best, cost-effective solutions to implement this. We also need a similar overall guide to do this. Thanks in advance.

For determining orientation dual antenna configurations with a separation >1m is suggested, more separation helps get better resolution.

Hi clive,
Isn’t it possible to achieve heading accuracy using a static base station? Can anyone suggest sources to set up a cost-effective GNSS RTK system with centimetre-level accuracy for a static base and moving rover to work without internet, including heading accuracy? Thanks in advance.

Define Cost Effective…
With a single antenna vs a base station you’d need to be moving to determine direction of travel. You could do a slight rotation with an off-center antenna to orientate. Or perhaps switch between a pair of antennas and calculate positions for each and map that against your understanding of them within the vehicle geometry.
You could use IMU or compass which you discipline against direction of motion whilst moving.

1 Like

Thanks for your reply. Setting up two antennas will be efficient for us. We are looking for low-cost products and setup guides available in the market.

I believe the low-cost route will be to use a consumer Autopilot (PixHawk, etc) and software such as Mission Planner with Rover Firmware. You have lots of options in this space in addition to the two I mention.

It’s likely that your selected AutoPilot can meet your actual design criteria with a single antenna RTK Rover (not a moving base) with it’s IMU/compass that @clive1 mentioned. A Moving Base Setup will significantly increase the heading accuracy when still or at slow speeds, but it adds complexity (which is fine, if you need it).

The catch is the common phrase “centimeter-level accuracy”.
Do you actually need 1cm accuracy, or could 1cm short-term repeatability suffice?
Because a true 1cm accurate position is hard to actually accomplish (even harder to prove), especially in motion.

You could start by researching the latest AutoPilots for one that could easily interface with a Sparkfun RTK product as the first step. Look for Moving Base optionally.

Then you can decide on your RTK correction source. If your Project Location will have access to the Internet (WiFi, Cellular), then you can use various Network Corrections and a Local Static Base isn’t required. If no internet access is available, then you will need to use a Local Base Station to transmit RTCM corrections to your Rover, or use L-Band (satellite) Corrections.

1 Like

Yes, mechanically very challenging, sort of situation where you’ll over-promise, and under-deliver, and all the non-technical people will get mad.

Vehicle is unlikely to be on a flat/level surface, so will pitch-roll-yaw, moving the antenna many cm vs notional point on the ground. All the mechanical linkages in the steering, and the time it takes for that to move in reaction to a correction in direction, with the vehicle having to move to turn.

1 Like

I’m using an F9P base and rover with LoRa serial for corrections. I’m able to stay in RTK Fixed most of the time. You could make it more economical with a cheaper base. It only needs to report corrections every couple of seconds.

The best way to get an accurate heading are the two GPS antennas at about a meter apart. I THINK but have not confirmed that you may be able to get the same accuracy with a dead-reckoning chip. That is only a few dollars more than the F9P alone.

This is all still pushing around $600. Not sure if that is cost effective.

my $0.02

Jay

We’re using an Intel NUC to process data from the IMU sensor, GNSS-RTK board, and encoder after the rover operates. For motor interfacing, we use an Arduino or ESP32. Since we have the NUC, I believe we do not require an autopilot, as the commands are sent directly from the NUC. However, if there is a specific use case for an autopilot, please let us know.

Additionally, could you clarify what you mean by ‘1 cm short-term repeatability’? What I meant by ‘1 cm accuracy’ is that if I get the position from the RTK, the difference between the derived location and the actual position should be a maximum of 1 cm, if possible, or at least within 10 cm.

Thanks in advance.

Our initial idea was similar to your setup. Could you let us know the level of accuracy your setup provides? Although the ZED-F9P mentions heading accuracy and centimeter-level accuracy in datasheet, I would like to understand why you suggest using the dead reckoning board, which costs $35 more as a kit. Are there any specific use cases where it offers significant advantages? Your insights would be greatly appreciated. also any report or recording of your setup is greatly appreciable.
Thanks in advance
Rakesh

Dead-Reckoning would be the option if you want one antenna/receciver. The F9R needs to move at some speed and do maneuvers to calibrated/fuse the sensors.
If you have a two antenna situation, and an IMU already, you could use the GNSS Orientation to disipline the IMU direction/compass as you have long/consistent motion to track.
cm level accuracy is “best case” with ideal conditions, I don’t think I’d be promising better than +/- 5cm, because that will simply become a stick you’re beaten with…
Also consider the volume of space your machine takes up, and where those notional 8 corners are as it pitches, rolls and yaws in 3D space.

Consider also that you’re moving, and the position is from where you WHERE when the GNSS took the measurements, as indicated by the time-of-day, this is apt to be 100’s milliseconds prior to the current time.

You field traversal, perhaps look at what AgOpen is doing, and the platforms it runs on.

Rakesh16 -
Clive1 is the expert here. I’m just now after experimenting understanding what exactly he means.

Specifically, These two GPSs have to communicate constantly with RTCM messages to stay in an RTK fixed mode. When in that mode, you do have very close to centemeter accuracy. If you move one meter forward and your GPS float is one centemeter to the left, you have about a half a degree error. HOWEVER, if the single antenna GPS falls out of RTK mode and your float is a meter to the left, then as you know you would have 45 degree error. Totally unacceptable for most applications.

If I understand dead reckonning correctly, it uses accelerometers to measure yaw so it would know you haven’t veered off course and your still tracking appropriately.

The second solution is if you have a two antenna solution and both fall out of RTK fixed, they would still be in relative alignment with eachother. So you could still get an accurate heading. The problem I see is if one is in RTK and the other isn’t, then you are no longer relatively aligned and the above situation persists. That is why my presumption is that dead reckoning would be the more consistent solution.

I’m sure there are inacuracies in my explanation. Others can chip in.
Jay

@Rakesh16 , since you’ve confirmed the requirement of 1cm accuracy, and also stated no internet access would be available at the project site - you’ll need to install and setup a Local Base Station.
To commission your Local Base with accurate coordinates, you will need to post process hours of static data (on several different days) to verify your Base Position down to 0X mm. That requires precise/final Orbit Data which is available 2 weeks after the survey date (for CSRS-PPP as an example). I wasn’t kidding when I mentioned " a true 1cm accurate position is hard to actually accomplish (even harder to prove)"

Now, you have a Local Base that can provide real-time corrections to your vehicle.
A single RTK Rover has the ability to provide decent heading information while it’s in motion.
If you want/need the best heading info, then a Moving Base is the solution.

A Moving Base (or Moving BaseLine) requires an additional receiver and antenna (it’s Rover) on the vehicle.
The Moving Base receives corrections from your Local Base (Static).
The Moving Base then “forwards” the RTCM corrections to the Rover.
The Moving Base is responsible for the precise location, and it uses the Rover for precise heading.

The Application Note that @clive1 linked to in his first post explains everything about the Moving Base application.

Note: Many/most autonomous vehicles perform just fine as a regular RTK Rover.

My suspicion is that you might be fine with the vehicle being able to return to a location within 1cm (precision), but that’s entirely different than KNOWING the vehicle’s absolute position to within 1cm accuracy.

There is a huge range of time & complexity required:
Spanning from just a single network RTK Rover that takes a 30 seconds to startup from a cold start… to 2-3 weeks to establish a <1cm Local Base + 2 other receivers for a Moving Base.

Your own project specification decides where you land in that range, which is why I previously said "The catch is the common phrase “centimeter-level accuracy.”

For the accuracy, you have to clarify what you want:

  • absolute accuracy i.e. each time I go to the same point, I want to be within the same world coordinates (or coordinates attached to local continental plate)
  • relative accuracy i.e. I want to be able to go to the same place any time later
  • short-term accuracy i.e. I want to be able to go to the same place within 10 mn

Stated accuracy of F9P and other dual frequency RTK devices is about 1 cm + 1 ppm of baseline length (distance between base and rover). So, if you install your base near the measurement area (<<10 km), you will get at least 1 cm accuracy. If you consider accuracy on base positioning itself and combining uncertainties, you will get something like:

  • absolute accuracy : 2-3 cm
  • relative accuracy : 1,4 cm (sqrt(2)x1cm)
  • short-term : should be < 1cm because error sources don’t change very fast (atmospheric disturbance, wrong satellites positions…).
2 Likes

As @clive1 said is it need to set 2 antennas in [ZED-F9P GNSS module] (SparkFun GPS-RTK-SMA Kit - KIT-18292 - SparkFun Electronics) to get minimum heading accuracy. We are currently using the ZED-F9P GNSS module and a LoRa module for transmitting base-to-rover corrections. As I mentioned earlier, we are also using an IMU sensor (MPU9250 and BMP280 module). Our system integrates this with an Intel NUC, which fuses the IMU data, GNSS RTK data, and encoder data to form the feedback system.

Could you clarify if the setup described would mitigate the errors you mentioned earlier? Specifically, we are aiming for a minimum of 10 cm accuracy even more accurate is okay for us.

Additionally, I would appreciate clarification from @clive1 or @Eric_S or @rftop , or anyone on the forum who has worked with a similar configuration:

  1. Will this setup reliably achieve at least 10 cm positional accuracy?
  2. What level of heading accuracy can I expect from this system?

Looking forward to your insights.

as mention in [ZED-F9P GNSS module] (SparkFun GPS-RTK-SMA Kit - KIT-18292 - SparkFun Electronics) will it be able to get 0.3 degree accuracy in rtk mode or what are all the additional products need to get atleast of 10 cm accuracy other than products which i mentioned above

yeah we need absolute accuracy and reduced our accuracy requirements to 10 cm till how many kilometres does it give such accuracy