System Identification and Robust Control of Farm Vehicles Using CDGPS

Automatic control of agricultural vehicles has been a research goal for many years. Previous attempts have failed largely due to sensor limitations. With the advent of modern GPS receivers, a single low-cost sensor has been synthesized in which centimeter-level position and attitude measurements of the vehicle state are available using Carrier-Phase Differential Global Positioning System (CDGPS). Previous research at Stanford University has demonstrated control of a Deere 7800 tractor at low speed without implements. Utilizing the Observer/Kalman Filter Identification method, input-output measurements were combined to form a central model of the farm tractor for all speed and implement combinations. A linear quadratic regulator (LQR) controller was designed utilizing the central model and experiments were performed to measure the achieved performance. Tight line-following was achieved that surpasses the capability of human drivers—at a speed of 1.75 m/s (4 mph), standard deviations for tractor with any implement were less than 7 cm. (3 in.) and with no implement, the standard deviation at this speed is less than 5 cm. (2 in.). Tractor automatic control performance improves as tractor velocity decreases, but remains excellent within the usable range—standard deviations remain at less than 11 cm. (4 in.) for all implement and speed combinations.