On November 19, China’s Ministry of Agriculture and Rural Affairs posted a notice dated November 9 and numbered MIIT Joint Office Letter [2026] No. 541. The body contained just three sentences: 55 typical application scenarios for agricultural robots had been selected; local authorities should strengthen policy support and promotion for the chosen scenarios; and applications should continue to expand. All 55 names appeared in an attachment.

The attachment is worth reading line by line. It is the result of two ministries spending five months screening submissions from provincial authorities, and it offers the clearest map yet of where agricultural robots are actually being deployed in China.

Tractors Already Took Care of Plowing

Of the 55 scenarios, 42 involve crop production, nine livestock farming and four primary processing of agricultural products. By stage, field management accounts for the largest share at 20, followed by harvesting at 11, animal husbandry at nine and seed and seedling production at seven. Planting, primary processing, and storage and transport each have two, while land preparation and end-to-end operations have one apiece.

Field management is the most crowded category, accounting for more than a third of the total. The distribution is easy to explain. Weeding, fertilizing, spraying and field patrols are among the most frequently repeated jobs over the course of a growing season. Each task is not especially demanding, but the total workload is enormous and spread across a long season. They sit squarely in the category of work that still requires people but little specialized skill. The logic behind the seven seed and seedling production scenarios is the opposite: four take place in greenhouses, covering melon grafting, rice seedling cultivation, solanaceous vegetable seedlings and sprout production. Light, temperature, humidity and seedbed positions are all controllable, making these among the easiest environments for robots to operate reliably. Neither extreme is surprising. The real surprise lies in the middle: land preparation received only one slot.

Another dividing line, absent from the table headings, is terrain. Five scenario names explicitly mention hilly and mountainous areas. Based on the locations of their lead organizations, two are in Sichuan, two in Chongqing and one in Zhejiang, all involving sloped land. There are few remaining gaps for large agricultural machinery on the plains. The real gap is where tractors cannot go.

Seven Provinces Report the Same Problem

Of the 20 field-management scenarios, seven involve weeding or mowing: smart crop protection and mowing; integrated unmanned weeding, fertilizing and spraying; smart row-following weeding for soybeans; integrated smart orchard inspection and weeding; smart orchard weeding in hilly and mountainous areas; and two scenarios both titled “Smart Laser-Weeding Robot Application Scenario.”

The lead organizations behind these seven scenarios are based in Shanghai, Shaanxi, Guangdong, Sichuan, Chongqing, Heilongjiang and Inner Mongolia, respectively. All come from different provincial-level regions, with no region submitting twice. Under Document No. 269, each province could generally recommend no more than five projects for each application stage. These were not repeated approvals for the same project, but seven separate applications that all ultimately received official endorsement.

The two identically named scenarios are also revealing. One lead organization is in Harbin and the other in Inner Mongolia, yet their titles are identical word for word. The same description appearing in two different parts of northern China suggests laser weeding is no longer a proprietary approach pursued by a single company. At least two organizations have advanced it far enough to win provincial recommendations and expert approval.

When seven provinces submit projects for the same stage, it suggests the economics are strongest there. Weeding relies entirely on manual labor, is priced by acreage and produces results that are immediately visible, making acceptance criteria much clearer than for fruit picking. It also shows that no single technical approach has prevailed. Mechanical cutting, row-following spraying and laser burning all appear on the same list, with none displacing the others.

Combines Already Dominate Staple-Grain Harvesting

The 11 harvesting scenarios comprise two for greenhouse tomatoes, two for tea, two for citrus, one for apples, one for lychees, two for button mushrooms and one broadly described as major fresh-eating fruits.

Not one involves rice, wheat or corn. Robots have little room to enter a job that machines can finish in minutes per mu, and there is scant scope for further gains. The category missing from the list is precisely the one with the highest mechanization rate.

The remaining ten all address the same challenge: removing produce without damaging it. Tomatoes must retain their stems, tea must be picked as one bud with one leaf, button mushrooms must be harvested without soil attached, and lychees cannot have their skins broken. Determining ripeness, finding produce behind foliage and reaching it without causing damage are three things combine harvesters do not do. These 55 scenarios are testing dexterity, not strength.

Button mushrooms account for two scenarios, led by organizations in Beijing and Anhui. Mushroom houses are indoors, with controllable lighting and fixed growing positions, making them the agricultural production lines closest to factories. It is unsurprising that two projects made the list. By contrast, apples and lychees have only one each. The more a crop depends on natural conditions, the fewer projects are able to establish a foothold.

Every Project on the List Has Already Entered the Field

A review of all 55 scenario names finds no mention of humanoids. The only one containing the term “embodied” is a heavy-duty embodied robot for open fields and orchards, led by a Shenzhen company.

Three days earlier, the Ministry of Industry and Information Technology had released five industry standards for humanoid robots. One of them, “Humanoid Robots: Classification of Intelligent Capabilities,” ranks capabilities from L0 to L5 across four dimensions: perception and cognition, decision-making and learning, execution performance, and collaborative interaction. Humanoid-robot stocks rebounded today, with media reports attributing the move to those five standards rather than this list. The standards define a scale for measuring intelligence, but not one humanoid appears on the list.

The reason appears in the application requirements of Document No. 269: each project must have already been used in at least one real agricultural production scenario. That condition defines the list. It is a report card, not a proposal.

The same document also stipulates the application structure: a robot developer must lead a joint submission with agricultural service organizations, family farms or agricultural companies, with no more than five participating organizations in principle. Each row in the attachment therefore lists two groups: those that build the machines and those that use them. Cooperatives, family farms, agricultural machinery extension stations and county-level agricultural technology centers fill the columns for partner organizations. This also explains why 46 of the 55 lead organizations are companies, while only nine are universities or research institutes. One institution, the Intelligent Equipment Technology Research Center of the Beijing Academy of Agriculture and Forestry Sciences, leads two scenarios. Companies, not laboratories, are putting these machines into the fields.

The Notice Does Not Mention Funding

Document No. 541 specifies no funding amount, subsidy rate or procurement volume. The most concrete line in Document No. 269 says local industry and information technology authorities and agriculture and rural affairs departments are encouraged to provide policy support and prioritize the promotion of selected scenarios. Words such as “encouraged” and “in principle” do not amount to commitments.

The process itself was lengthy. The document was issued on June 9, provincial submissions closed on August 14, the final list was dated November 9 and it was posted online on November 19, a span of 163 days. Projects passed through five stages: company applications, recommendations from relevant organizations, expert review, consultation and public notice. There are reasons for moving slowly, but the delay also means projects may be unchanged from the time they were submitted to the time the list was published.

The real switch is not in either notice, but in the catalog for agricultural machinery purchase and application subsidies. Whether these projects enter that catalog will determine whether the 55 scenarios translate into recognition or orders.

Price remains an unavoidable barrier. According to a January report by China Jiangsu Net, domestically made single-arm harvesting robots cost 150,000 to 250,000 yuan each, while multi-arm models cost 500,000 to 800,000 yuan. Imported models cost another 30% to 50% more, and annual maintenance runs from 10,000 to 20,000 yuan. In the same report, a company estimated that a 100-mu tomato farm could save 150,000 yuan a year. The economics work for a 100-mu commercial farm, but not for a 10-mu household-contracted plot. Behind most of the 55 scenarios is an operator supported either by subsidies or sufficient scale. That helps explain both why they made the list and why they will be difficult to replicate.

What to Watch