Spelling suggestions: "subject:"farm aautomation"" "subject:"farm andautomation""
1 |
Robotic cotton harvesting with a multi-finger end-effector: Research, design, development, testing, and evaluationGharakhani, Hussein 12 May 2023 (has links) (PDF)
Cotton is harvested with large and heavy machines that are very efficient but have some disadvantages. They can harvest the crop only once at the end of the growing season. Since cotton bolls do not mature uniformly, the early opened bolls expose their fiber to weather for extended periods, reducing lint quality. In addition, the machines can also compact the soil, reducing water and fertilizer usage efficiencies and crop yields in the following years. Robotic cotton harvesting offers a promising solution to these issues. Smaller robotic harvesters could go to the field multiple times during the season to pick cotton bolls as soon as they open. Such harvesters could be lightweight, minimizing the risk of soil compaction. This dissertation research includes designing an end-effector for robotic cotton harvesting, designing a robotic platform and integrating the custom-designed end-effector, and developing multiple manipulation control algorithms. The robotic platform has a 3-DOF (degrees of freedom) manipulator and a ZED 2i stereo camera. The robot was tested under lab and field conditions to evaluate its performance in object detection, localization, and picking.
The tests proved that manipulating the arm while picking a boll increased the picking ratio – the weight of the picked seed cotton over the whole weight of the seed cotton that the robot attempted to pick – by up to 23%. However, it increased the cycle time. Therefore, the control algorithm was improved to a closed-loop system to touch just the unpicked areas of a boll. The best control algorithm, i.e., I-FMW (improved-feedback-based manipulation while picking), could achieve a 72.0% picking ratio with a cycle time of 8.8 s during lab tests.
The field tests were conducted to find the contribution of three main systems (detection, localization, and picking) to the losses. The tests showed that detection, localization, and picking subsystems could achieve performance of 78.1%, 70.0%, and 83.1% respectively. Therefore, detection and localization systems must be improved. Utilizing better sensors, modifying detection and localization algorithms, adding the boll orientation information, and controlling illumination conditions as much as possible would improve the picking performance and make the robot a step closer to a commercial product.
|
Page generated in 0.0987 seconds