Modern agricultural robots rely on knowing exactly where they are.
In open fields, that’s often straightforward. Satellites provide positioning data that helps robots move through crops, complete tasks and return to the right location.
But not every farm offers a clear view of the sky.
Dense orchards, vineyards, greenhouses and even tunnels can weaken or completely block GNSS (Global Navigation Satellite System) signals, leaving robots without one of their most important navigation tools. If they cannot accurately determine their position, even simple tasks like crop inspection become difficult.
Researchers from the National Research Council of Italy (CNR) have been exploring a different approach—one that allows robots to navigate without relying on GPS at all.
Using the environment as a guide
Rather than asking, “Where am I on the map?”, the robot asks a simpler question:
“Where am I in relation to the wall beside me?”
Using an RGB-D camera, which captures both colour images and depth information, the robot continuously analyses nearby vertical surfaces. These could be vineyard rows, greenhouse structures, orchard canopies, tunnels or even cave walls. By estimating the position and orientation of these surfaces in real time, the robot can calculate its own location and maintain a safe, consistent path alongside them.
Unlike many modern navigation systems, this method does not depend on machine learning models that require extensive training data. Instead, it uses geometric calculations to interpret the environment directly, making the system easier to deploy in new locations without collecting and labelling large datasets beforehand.
Designed for the real world
Agricultural environments rarely behave like laboratories.
Leaves move with the wind, lighting changes throughout the day and sensor measurements are never perfectly clean. Any navigation system must be able to cope with these uncertainties while keeping the robot on course.
To test the approach, the researchers first evaluated the system under controlled laboratory conditions before moving to real vineyard trials. The robot successfully followed vineyard rows using only onboard sensing, while the researchers also compared different methods for estimating surrounding surfaces.
Among the tested approaches, a Least Squares algorithm proved particularly well suited for real-time operation, achieving accurate positioning while requiring only fractions of a millisecond to process each update. The robot was able to reach its intended path in under 20 seconds and continue navigating reliably despite measurement noise and changing outdoor conditions.
Why it matters
As agricultural robots become more autonomous, reliable navigation will be just as important as robotic arms, cameras or AI models.
Many agricultural tasks take place exactly where satellite signals are weakest—beneath tree canopies, between vineyard rows or inside protected growing environments. Robots operating in these conditions need alternatives that don’t depend on external positioning systems.
This research demonstrates that, in many cases, the surrounding environment itself can become the robot’s guide. By interpreting the geometry of the world around it rather than relying solely on GPS, agricultural robots can continue working safely and accurately, bringing autonomous farming one step closer to operating reliably wherever crops are grown.