This paper reports on an operation framework for autonomous rice harvesting. We developed an integrated algorithm for robotic operation and cooperation with farmworkers to automate each subsection of the harvesting and unloading process and of the processes that bridge them (homing and restarting). The algorithm was installed into a head‐feeding combine robot. The robot followed a target path based on its absolute position and orientation, planning a counterclockwise spiral path in a rectangular paddy field, and returned to a position close to a farm road when its grain tank was filled to a specified level. The grain unloading operation was automated using a machine vision system. As the restarting process (return to harvesting) was also automated, the combine robot was able to harvest a rectangular field autonomously by cyclically repeating the harvesting, homing, unloading, and restarting operations. Under field conditions, the robot was able to follow the target path within tolerable lateral and azimuth errors while harvesting rice successfully, and to unload the harvested grain into a wagon without spillage. The root mean square error of the lateral and azimuth errors during harvesting were 0.04 m and 2.6°, respectively. In the homing operation, the robot returned to a given line within ± 0.1 m and aligned its heading to the direction of the line within ± 4°. The robot recognized the arbitrarily parked wagon and positioned its auger spout at the target point with a respective horizontal and vertical accuracy of ± 0.2 m and ± 0.3 m. Harvesting time accounted for 50%–60% of the entire robotic operation. Homing scheduling and dispatch control for the wagon were found to be of importance for developing a more efficient robotic operation.

No abstract

No abstract

No abstract

This study proposes a method for detection of uncut crop edges using multiple sensors to provide accurate data for the autonomous guidance systems of head-feeding combine harvesters widely used in the paddy fields of Japan for harvesting rice. The proposed method utilizes navigation sensors, such as a real-time kinematic global positioning system (RTK-GPS), GPS compass, and laser range finder (LRF), to generate a three-dimensional map of the terrain to be harvested at a processing speed of 35 ms and obtain the crop height. Furthermore, it can simultaneously detect the uncut crop edges by RANdom SAmple Consensus (RANSAC). The average of the lateral offset value and crop height of the uncut crop edge detected by the proposed method were 0.154 m and 0.537 m, respectively.