NeuroAI NEUROAINEUROAI.SITE
China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories
ESC

From five days to one: China's multi-arm apple picker takes on the harvest

A Beijing-built four-arm apple picker cleared one mu (亩) of orchard in a day where workers needed five — a field test that shows orchard robotics moving from demo to daytime work.

2026-11-13 · 822 words · NeuroAI
From five days to one: China's multi-arm apple picker takes on the harvest

On a hillside apple orchard outside Kunming, a machine on wheels rolled slowly down a row of dwarf trees, reaching with four arms toward fruit that most people would need a ladder and steady hands to pick. Around it, the usual crew of workers was nowhere to be seen.

This was not a trade-show stunt. On November 1, 2026, the fourth-generation apple-picking robot developed by the Intelligent Equipment Research Center at the Beijing Academy of Agriculture and Forestry Sciences (北京市农林科学院智能装备技术研究中心) ran a live demonstration at the Tuanjiexin apple estate in Kunming's Xishan District. It is the clearest sign yet that fruit harvesting — long considered too delicate and too unstructured for machines — is being pulled into the era of embodied intelligence (具身智能).

One acre, one day

The number that matters to any orchard owner is simple. Picking one mu (亩, about 0.067 hectares) of apples by hand takes roughly five days of labor. The four-arm machine, working alongside an autonomous navigation transport cart, finished the same area in a single day.

The system is designed as a closed loop: recognize (识别), pick (采摘), stack (码放), and transport (转运). Flexible end-effectors wrapped in foam and food-grade silicone cradle each apple; a soft conveyor buffers the fruit into crates. In its standard test conditions, the robot ran eight to ten hours continuously and kept the comprehensive damage rate below 10 percent.

How it actually picks

Recognition is the hard part. In standardized dwarf orchards (矮化密植园), the machine identified mature apples at a 92.94 percent rate and located each fruit within ±5 millimeters. Four arms working together reached about 1.96 times the efficiency of a single arm — the gain that turns a curiosity into a tool.

The operator controls it from a tablet. According to the demonstration team, it has already been adapted to common Chinese varieties such as Ruixue (瑞雪), Ruiyang (瑞阳), and Yanfu No. 3 (烟富3号), and covers more than 85 percent of the orchard's working conditions.

The machine did not appear from nowhere. Earlier in 2026 it had already been shown in the main apple regions of Shandong, Shanxi, and Shaanxi, with a cumulative demonstration area reported at around 50,000 mu (5万亩). The Kunming site is the first field run in the hilly red-soil terrain of central Yunnan (滇中丘陵), where the city began building an apple-harvesting robot application scenario in 2025.

Why orchards are betting on it

China is the world's largest apple producer, and the labor that used to be cheap and plentiful is no longer either. An aging rural workforce and rising wage expectations make the harvest the single most labor-intensive and time-critical step in the orchard year. A machine that is five times faster per area, runs all day, and does not call in sick during peak season is exactly the kind of leverage growers want.

According to Li An, chief scientist at BrainNet (脑机网), the move from single-arm lab demos to navigation-equipped, multi-arm field machines shows embodied intelligence (具身智能) leaving the showcase and entering production lines where the economic case — not the demo video — decides whether it stays.

The apple machine is not alone. Through 2026, Chinese labs and firms demonstrated strawberry pickers with soft, tentacle-like grippers, grape-specific robots built for the Malu vineyards near Shanghai, and a bilateral human-like tomato picker that reported a 90 percent success rate in Chongming greenhouse trials. In January 2026, a national key laboratory for intelligent harvesting robots (农业农村部智能采收机器人重点实验室) was approved, a signal that fruit picking is now a funded national priority rather than a scattered research hobby.

Honest limitations

This is still a demonstration, not a commercial product on every hillside. A few caveats matter:

  • The 92.94 percent recognition figure comes from controlled dwarf orchards. Real orchards add overlapping branches, uneven light, disease, and variable tree shapes that will push the miss rate higher.
  • An under-10-percent damage rate is a real improvement, but it still means lost fruit and, more importantly, fruit that may not survive the shelf.
  • "Covers 85 percent of conditions" means the other 15 percent — unusual layouts, steep slopes, mixed varieties — is out of scope for now.
  • Picking is only one harvest task. Pruning, thinning, and quality grading are not solved here, and the transport integration is early.
  • The throughput and damage numbers come from a single research-institute demonstration event. Independent, multi-season, multi-orchard validation has not yet been published.

What readers can do now

  • If you run an orchard, request a demonstration during your own harvest window and measure damage rate and labor substitution on your varieties and row layout — not on the vendor's test block.
  • If you follow the market, watch the Shandong–Shaanxi–Beijing demonstration expansions and wait for a published per-unit price in RMB before committing capital; robotics economics live or die on utilization.
  • If you write policy or invest, separate demo throughput from season-long reliability, and ask for multi-year, multi-site data before treating "five times faster" as a baseline.

Related coverage

Cite this article

NeuroAI (脑机网), "From five days to one: China's multi-arm apple picker takes on the harvest," 13 November 2026, https://neuroai.site/a/na-robot-apple-picking-robot

Machine-readable: BibTeX · RIS · JSON · Markdown

More in “Humanoids & Robotics” → · Back to home · Markdown version

China AI, reported clearly
1,300+ verified stories on models, robotics, BCI and the business of AI.
Browse stories