Vol. 12, Issue 7, Published 4/7/2021
Over the last several years, there has been a tremendous amount of research activity towards the development of autonomous agriculture vehicles. A quick internet search will reveal over 50 companies or university research groups working in this space. A question I get often from groups developing such platforms is “What is a good agricultural application for our lightweight “robot”?”. It’s a great question, and for Arizona vegetable production, it’s also one that I’m not sure I have a satisfying answer for.
The calls I get regarding autonomous robots are mostly related to automated weeding applications. Automated weeding machines are commercially available, but their adoption has been limited not because of labor costs for tractor operation, rather it is the lack of the development of a functional and cost-effective means for identifying and removing weeds.
For decades, researchers have been attempting to develop sensing systems that are able to reliably detect weeds. Techniques such as 2-D and 3-D color imaging, x-rays, hyperspectral sensing and artificial intelligence have been tried (Slaughter, 2014; Bender et al., 2020). The best performing systems provide about 96% accuracy, meaning that 4% of the crops plants are identified as weeds and would be destroyed by the weeder. For high value vegetable crops like lettuce with gross revenues of roughly $10,000 per acre, killing 4% of the crop equates to $400 per acre of losses. Economically, this does not make sense as hand weeding labor costs are typically $300 per acre or less. The other main issue is that current automated weeding technologies are not highly precise and provide only partial control. Our studies with these types of machines have shown that these systems remove only about 1/3rd of the in-row weeds (Lati, et al., 2016) and a follow up hand weeding operation is often necessary. To be highly cost effective, elimination of the hand weeding step is needed.
In short, my recommendation to research groups asking about applications for autonomous robots is that their time and technical skills would best be served developing reliable crop/weed differentiation systems and a technique to remove a very high percentage of weeds.
References
Bender, A., Whelan, B. & Sukkarieh, S. 2020. A high‐resolution, multimodal data set for agricultural robotics: A Ladybird's‐eye view of Brassica. J. Field Robotics. 37(1): 73-96.
Lati, R.N, Siemens, M.C., Rachuy, J.S. & Fennimore, S.A. (2016). Intrarow Weed Removal in Broccoli and Transplanted Lettuce with an Intelligent Cultivator. Weed Technology, 30(3), 655-663.
Slaughter, D.C. The biological engineer: Sensing the difference between crops and weeds. Autonomous robotic weed control systems: A review. Computers and Electronics in Agriculture 61(2008): 63-78.
