Polybot's Tomato Robot Learns to Harvest by Watching Humans

Polybot, a Tübingen-based startup spun out of the ELLIS Institute and the Max Planck Institute for Intelligent Systems, is building a tomato-harvesting robot that learns its job by watching people work.

The machine has already left the laboratory and is operating in real greenhouse growing rows. The significance, in the company's framing, is that agricultural automation may be reaching the point where it becomes practical for a role with very high demands on precision.

Polybot's system is designed for fully autonomous harvesting of truss tomatoes — the kind sold still attached to the vine. Rather than engineers writing rules for every motion, the robot is trained end-to-end on human demonstrations.

Truss-tomato picking is nothing like handling standardized parts in a factory. Fruit must be grasped and placed gently, and the robot has to deal with the natural variation of a living plant rather than identical, uniform products moving along a conveyor.

Learning from the actions of human workers instead of being programmed step by step may allow the system to adapt more effectively to complex growing environments. For growers, that translates into support for one of the most repetitive and physically demanding jobs in food production.