For the last 6 years, small autonomous agricultural vehicles have been operating on Harper Adams University’s fields in Shropshire.  Starting with a single tractor on a single rectangular hectare (2.5 acres) and moving on to three tractors on 5 irregularly shaped fields covering over 30 hectares (75 acres).  Multiple crops have been grown; planting, tending, and harvesting with autonomous tractors and harvesters.  The fields are worked using a Controlled Traffic Farming system, which means that the routes taken for all farming operations in a season must be known prior to the start of the season.  Route plans are generated prior to entering the field and include the intended route, the desired speed for each route segment and whether the implement is in work on any particular segment.  An analysis of the route plan yields information about the expected distance travelled, the proportion of time in work and the overall time predicted for the operation.

Over the course of the Hands Free Farm project hundreds of hours of telemetry has been gathered operating four different vehicles on almost 100 different routes over 5 different fields. The telemetry data includes information about the time, location, speed and operational status of the vehicle. An analysis of the telemetry data yields information about the time in work and the overall time taken for the operation. By comparing the analysis of the telemetry data and the route-plans, it is possible to determine how accurate the route-plans are at predicting the real world operation times. There are several causes for the differences between the predicted and the actual operation times including the local geodetic height, field topography, vehicle limitations, and the physical inertia of the vehicle and implement.  In all cases, real-world effects result in an increase in actual time taken over that predicted by the route-plan analysis.  Further work can be done to integrate these real world effects into the analysis process and system improvements can be made to improve real-world performance and reduce the discrepancy between predicted and actual work rates.

Having established the reliability of the route-plans at putting a lower bound on field work times, several benefits to precision agriculture can be realised:  When planning future cropping cycles, management strategies can use the data from more accurate predictions of field work to improve decision making.  When generating route-plans for fields, machine learning techniques can be applied to optimise routes and find more efficient methods for farming oddly shaped fields.  Combined with analysis of historical yield data, route-plans could be generated for improved resource use efficiency by setting aside unproductive areas within a field for improved biodiversity while also optimising routing and improving the profitability of the field as a whole.