I’m exploring the idea of using GPS sensor exclusively to guide a mobile robot along predefined line patterns on sports fields (such as soccer fields, tracks, or basketball courts). The goal is to rely solely on GPS data for navigation—without the use of cameras, vision systems, or any additional sensors. Is it possible?
My main objective here is to investigate the weaknesses and limitations of using only GPS for precise navigation. This exploration could serve as a foundation for later expansion by incorporating multiple sensors or technologies, but for now, I want to focus on how far GPS alone can take us.
What are the specific goals you want to achieve? E.g. are you going to paint the lines on the field, or do you have some slack? Do you need mm, cm, m accuracy? If you go for an RTK GNSS solution, you should reach <m or <10cm accuracy if you have clear view of the sky.
I want to investigate how accurately a mobile robot can navigate the lines, and when its navigation based GPS is perfect when no noise is introduced, then that can be considered the “ground truth.” However, in another scenario, GPS sensor will be introduced with noise due to factors like line of sight issues or weak signals, which will affect the accuracy of the robot’s position. The weakness of GPS due to this noise or other factors can be mitigated later through techniques like Extended Kalman Filtering (EKF) or other techniques methods through sensor fusion. The key challenge lies in dealing with the imperfections and noise from GPS and improving accuracy as the system progresses.
I would definitely go for GNSS and not GPS only. Galileo offers the best precision and Beidou has the most satellites. GLONASS is not that much useful these days because of its old technology. All constellations if possible would be the best, the receiver should itself select what is currently the best combination.
You will definitely want at least 2-frequency receiver, better 3-frequency (this allows the receiver to cancel out ionosphere effects).
You will also need RTK corrections. If you use your own RTK base in the corner of the field, it should give you the precision of a few mm. But you need to figure out how will the base communicate with the rover (usually for these shorter distances, LORA is sufficient).
Last, for actual navigation, you also need the heading of the robot. Using magnetometer-based compass is very imprecise. Better solution is a 2-headed GNSS receiver with antennas as far apart as possible. Then you get heading of the robot with ~2-3° accuracy. Septentrio Mosaic-H is one such product that is 2-frequency and has 2 antennas.
Last, the output of most GNSS receivers is 5 or 10 Hz max. You should consider if that’s enough for your task.
Any kind of EKF on top of just GNSS measurements does not make too much sense. It would only smoothen-out the sharp corners you need. It would be better to integrate at least wheel odometry and IMU.
@peci1 and @Wilco made an already good response to the issue, so I will just add my personal thought working with GPS. I think that the issue here is relying only and just only in one single sensor in order to achieve localization. If you have clear view of the sky you will probably be good to go with GPS-GNSS, but even as small as a minimum deviation might cause several issues finding the exact ground truth position of your robot. Moreover, you don’t achieve RTK signal in every single scenario, which obviously it is a problem if you require cm accuracy. Also, bare in mind that the lecture from a GPS is discrete, while other sources such as odometry is continous, so you probably will have to include any other sensor with an EKF as already pointed out
While the original question has been answered:) Agree GPS with dual antenna should provide a position solution under 0.5 m and using RTK (RTCM signal) lt 10 cm is possible. Information for correction stations can be found here for the US - NOAA CORS Network - National Geodetic Survey .
@Johanes_PS you could use a UBLOX F9R or two F9Ps with a moving base station configuration. Its possible to get CM level accuracy with our driver UBLOX DGNSS
I always however say to people never trust just one sensor - if you can have multiple independent sensors the more reliable your solution
Dont know what GPS youre using… but seems a lottle tight. Cloud coverage and other factors could derail that 0.5 m accuracy. I think you need another complementary sensor if you need very accurate positionnig. RTK could be. But i think you need another refrence point.
Cloud cover should have non-observable effect on GNSS signals (at least this is what guys from radio department say). If you have multifrequncy receivers and RTK with a closeby base, all atmospheric effects should be effectively cancelled out. The only imprecisions that remain are multipath reflections and receiver clock instability. Multipath from buildings should not be a big problem for a football field robot, but reflections from ground could be. We’ve figured GNSS antennas need to be at least 1 m above ground, otherwise they’re basically useless. Or you could buy the super expensive antennas with choke rings.