China's Ye Shu Group Wants a Coconut-Peeling Robot: 360 Coconuts Per Hour, Less Than 1% Damage

The Tender

Ye Shu Group, maker of China's iconic Ye Shu coconut milk (椰树牌椰汁), has issued a public tender seeking an automated coconut-peeling system with demanding specifications:

• Throughput: Minimum 360 coconuts per hour (6 per minute)
• Damage rate: Less than 1% (damaged = unusable for processing)
• Requirement: Fully automated, minimal human intervention

This is not a trivial engineering problem. It exposes the gap between AI robotics demos and real-world factory floors.

Why Coconuts Are Hard

Unlike manufactured objects, coconuts are profoundly irregular:

Physical Challenges

• Size variation: Coconuts range from 12-25cm in diameter
• Shape variation: Spherical to ovoid, often lopsided
• Shell thickness: Varies significantly by variety and age
• Weight: 1.5-2.5 kg each
• Fiber layer: The outer husk is tough, fibrous, and grippy — hard for robotic manipulators
• Water content: The liquid inside shifts the center of gravity during processing

Industrial Challenges

• Tropical environment: High humidity, temperature — bad for sensitive electronics
• Continuous operation: 360/hr means 8,640/day in a factory running 24 hours
• Maintenance: Fibrous debris clogs mechanisms
• Cost: Must be cheaper than manual labor to justify investment

The Human Baseline

Currently, coconut peeling is done manually. A skilled worker can process roughly 80-120 coconuts per hour with high accuracy. The robot target of 360/hr represents a 3-4x productivity improvement — ambitious but not impossible for a well-engineered system.

Manual labor costs in Hainan Province (China's coconut hub) are rising, making automation increasingly attractive.

Technical Approaches

Computer Vision

The robot must first scan each coconut to determine:

• Size and shape for grip adjustment
• Orientation for optimal cutting approach
• Shell thickness estimation for blade depth control

Modern depth cameras (Intel RealSense, structured light) combined with ML classification can handle this.

Robotic Manipulation

The gripping problem is the hardest part:

• Soft grippers — silicone or pneumatic grippers that conform to irregular shapes
• Custom end effectors — designed specifically for coconut geometry
• Multi-stage processing — hold, stabilize, cut, peel, eject

Cutting/Peeling Mechanism

Options range from:

• Rotary blades — fast but risk of over-cutting into the meat
• Water jets — precise but slow and messy
• Laser cutting — expensive but potentially precise
• Mechanical peelers — traditional approach adapted for automation

The Broader Context

Ye Shu's tender is part of a larger trend in Chinese food processing automation:

China's food processing industry faces a perfect storm for automation: rising labor costs, aging workforce, quality consistency demands, and government subsidies for smart manufacturing.

The AI Robotics Reality Check

This tender is a useful reality check for the AI robotics industry:

• Demo vs. Production: Picking up a standardized object in a lab is one thing; processing thousands of irregular natural objects in a humid factory at speed is another
• The last 1%: Going from 95% to 99% accuracy typically costs 10x more than the first 95%
• Dirty environments: Real factories have debris, moisture, and variability that break lab setups

If someone delivers on this tender, it would be a meaningful demonstration of industrial robotics capabilities. If no one can meet the specs at a reasonable price, it reveals how far we still have to go.

Source: Zhihu Discussion