Abstract: Abstract: Target spraying can improve the utilization rate of the liquid for less environmental pollution, compared with traditional spraying. There are technical requirements for stable and reliable recognition, as well as accurate nozzle position solving in the spraying system. This research aims to develop and evaluate the accurate position-solving and error correction of nozzles for targeted spraying using a pre-designed field robot. The high accuracy of target spraying was achieved by plant protection machinery in unstructured field environments. The field robot of target spraying was mainly composed of electromagnetic nozzle, suspension, walking chassis, walking and, target spraying control system, as well as the global navigation satellite system. An Unmanned Aerial Vehicle (UAV) was used to collect the field information for the prescription map with the target spraying operation task. Specifically, the memory card was inserted with the prescription map information into the main board of the target spraying control system, and then to guide the robot for the target spraying. The robot was combined with the positioning and orientation data to solve the coordinates of each nozzle in real time during operation. Among them, the structural parameters of the robot were compared with the prescription diagram, in order to control the movement of the nozzle for target spraying. As such, the on-target spraying operation was implemented during the robot walking in a complex field. The position of the spray nozzles was solved to consider the errors originated from the production, installation, and movement of the components. The cumulative effect of error transmission between moving part was evaluated for each part of the error, compared with the robot kinematics. Among them, the end-to-end coupling error was transformed and described uniformly, and then decomposed and quantified at the end of the error transfer. The final coupling error was equivalently decomposed into the decomposition errors in six directions, including translation errors in three directions, and rotation errors around three axes. The error values were derived within the range of suspension motion under a combination of field measurements and theoretical calculations using Gaussian machine learning. The auto-regression learner established the correspondence between the length of the electric cylinder on the suspension and each error, thus enabling the prediction of the errors and the correction of the nozzle error solution model. The mean correction on the nozzle position of the solution model fully met the requirements of large field operations, compared with the commonly-used one. Finally, the corrected model of the nozzle position solution was deployed to the edge end. Leveling ground and field trials were conducted to verify the model. The results indicated that the accurate estimation of the robot's structural parameters was achieved using the Gaussian regression modelling. The average deviations were 4.3 and 1.3 mm for the relative height and relative distance between the nozzle and the positioning point, respectively. The mean plane error of the nozzle position solution was 8.5 mm. The longer the response distance of the nozzle from the target center was, the higher the target spraying accuracy and the better performance of the system were. The longer the response distance of the nozzle from the target center was, the higher the target spraying accuracy and the better the stability of the system were. Furthermore, 94.4%, 96.6%, and 99.4% of the samples were sprayed to the target with an accuracy ≤30 mm at 0, 15, and 30 cm target guidance distance, and the coefficients of variation were 0.010, 0.017, and 0.010, respectively, when the field travel speed was 1 m/s. Higher accuracy was achieved in the nozzle position solution and operational stability. The accurate calculation and correction of the nozzle position can be used for the precise control of the ground robot end-effector in the air-ground cooperative robot system.