A farmer who utilizes precision farming takes care of small areas within a field individually based on the varying needs that change across a field. Two good thing result from this kind of farming:
1) Production is cheaper
2) Damage to the environment is reduced since chemical application is reduced to only what is necessary
Precision farming doesn’t just utilize one technology. Tech includes GPS, GIS (Geographical Information System), monitors for yield, remote sensors, and technology that disperses products and varying rates.
GPS can be utilized for guiding farming vehicles. This kind of vehicle would show up on an already-installed monitor. A farmer could control the vehicle by utilizing GIS to decide which way to go, and determine where and which way the the vehicle is going according to the GPS unit attached to the vehicle. Using these technologies in tandem, a farmer can “drive” his farm equipment exactly where he wants it to go.
A GPS unit that is useful for such a use must be very accurate in order to actually reduce the amount of application made, and must be easy for a farmer to use – not to mention, economical. Extremely precise technology exists, but it’s usually too expensive and techie for someone whose expertise is NOT computers. The tech that is often used in farming equipment is instant kinematic positioning (DGPS RTK) that ensure accuracy because it self-corrects. This technology hopes to be low cost and show equipment at a certain location. Such technology was first made for ships at sea, but we hope to make them usable for land vehicles – especially farming vehicles. The first experiments with such technology were not satisfactory due to a Kalman filter that was inside the software. Such a problem is giving way to the development of new tests wherein a different algorithm will be developed to better the software. We hope the end result will be a low-cost way of receiver communication that enhances a vehicles ability to locate itself and determine its trajectory so that it can carry out farming duties – driven remotely.
What We’ve Tried
GPS receivers for this applications require, not only an high accuracy to ensure the reduction of input products, but even an easy and immediate way of use for farmers; without forgetting low costs. Obviously the technology to achieve high precision still exists but it is too expensive and difficult to use for not skilled people. Survey modality usually adopted in agricultural applications is real time kinematic positioning, DGPS RTK, which enable tohave a good accuracy by means of corrections received. In this experimentation the aim is to obtain a sub-metric accuracy using low cost receivers, which can provide only point positioning. These receivers have been developed for maritime navigation purposes; our aim is their optimization in order to apply them for land navigation in particular for farming activities. Some tests using these receivers were carried out, but results were not satisfying and probably the reason has to be assigned to the implementation of a Kalman filtering inside the receiver software. This is the starting point for a new project, at the moment still in progress, which aim to develop a new algorithm based on Kalman filter. Its purpose is to improve low cost receiver outputs in order to optimize trajectories and to reach needed accuracy in vehicle positioning during agricultural activities.
1) Test and Tools
Leica Geosystems were the brand of instruments that we used. Namely, their inexpensive receiver: TruRover Leica. It has an antenna for recpetion, 5 Hz tracking time, uses NMEA string format, and is unable to show positioning simultaneously or store the data. This hardware needs a separate computer to show its data. TruRover has been compared with Geodetic receivers (much higher technology), therefore making them good receivers to assess the TruRover’s abilities.
We used the GX1230 from Geodetic which can receive phase AND code frequencies. Static and kinematics tests were both done in an effort to mimic a real farming vehicle’s actions while simultaneously using bot GPS receivers. The TruRover and Geodetic Antenna were attached. They were both connected to the receiver and were 50 cm apart. During twenty mins., 3 static stops were done. The Geodetic receiver was set at ten degree cutoff angle and tracking time of one second. The test lasted around 2 hours. A second geodetic receiver used for pinpointing was put at the master station to process data.
2) What to do with the data
Static was processed to determine the locations of two master stations: Modena station (where the tests were done by INGV), and the ASI Telespazio-led Bologna station. Included in TruRover data is coordinates. Software to visualize these coordinates displays them and stores them. Such data is looked at alongside the data collected through two frequencies in the kinematic testing. Deciphering this data can give the trajectory of a vehicle. A special software, Leica Geo Office, was used to decipher that data. Considering the errors present in post processing, trajectories cannot be 100 percent accurate. Yet the data is accurate within centimeters – which is more than farming needs require. However, it’s impossible to really compare TruRover with the exact one since there is still a shift of 50 cm. After post-processing, a sort-of overlap is apparent to see the trajectory.
Results of this test did not please. Specifically, the receiver could not show an accurate curve. The estimate was bigger than the real curve. The theory is that the Kalman filter written into the software is causing this issue. Most likely, the current algorithm uses the current position to predict upcoming positions based on a consistent velocity and projected trajectory. Because of this, the vehicle’s real curves look like mistakes to the computer, which responds by “correcting” reality. The result is delayed curving, and a shift from the real location.
Yet, extremely accurate curves are not as necessary as extreme accuracy in straight lines, which is where most farmers do their farming. On the other hand, curves play a special role since curves at the end of the line lead to the following line – which must be parallel to the previous line. This is because it is essential to ensure that things are spread out on the field efficiently. It is the kinematic one that is the precise trajectory, which can be used to compare tracks from kinematic trajectories and pseudo-range trajectories.
The goal was 1 meter distance, but the end result was greater. Yet the result was less than ten meters. Comparing to parameters for statistics, the standard deviations and means are confirmed. Originally, we thought the Kalman filter would work better if it had more time to assess the situation. However, more time just made the situation worse.
Creating a New Algorithm
These discrepancies are probably because of the Kalman filter which wasn’t specifically made for agricultural use. So a new algorithm must be developed specifically to fit agriculturalists’ needs.
Right from the start, using Kalman equations was a problem. Two problems have been illuminated: how Kalman deals with velocity and acceleration. It treats both as if they are constant. These are Kalman issues.
What has been described from these test present huge issues for precision farming. Bad tracks result in wasting goods in addition to costing more and being worse for the environment. Now, a new algorithm based on Kalman has been created. This new algorithm will be a stripped-down version of the “auto-correcting” Kalman. This time, though, it will account for a farming vehicles natural movements and the predetermined path planned by the farmer and then works on constantly re-evaluating its prediction for where the vehicle will be going. This will take out any drifting curves and huge increases in changes to the satellite configurations. This research is still going on, though we have already implemented an algorithm that utilizes the constant velocity in straight lines and how a vehicle increases in speed on a curve. Since we didn’t have un-filtered data, it was difficult to make a good algorithm. TruRover data continues to be strained through a filtering system, and it’s impossible to get the raw data. That’s why we have used data from other receivers to create a new algorithm. What we’ve found out is that it’s crucial to create an algorithm that takes curve acceleration into account, and yet continue to look at the resulting data critically. Tests will take special note of covariance or measurement noise.
There are two steps that need to be taken now:
1) both noise and measurement matrices will weigh varying covariances
2) use unfiltered data with inexpensive and mono frequency receivers.
It is quite a challenge to use TruRover for precision farming, considering we only have raw data. Alternatives include:
– We must integrate the steering and odometer with the GPS system that already is capable of human remote control. That way, we could use that as the input for the Kalman filter.
– We need to make use of DGPS, differential positioning, and improve coordinates based on data by a Master station near the field.