Learning to Prune Branches in Modern Tree-Fruit Orchards

TL;DR

This paper introduces a robotic pruning system with a visuomotor controller trained in simulation, capable of guiding cutters in cluttered orchards using optical flow, achieving effective zero-shot transfer to real-world scenarios.

Contribution

It presents a novel closed-loop controller trained in orchard simulation that uses optical flow for pruning, eliminating the need for full 3D reconstruction.

Findings

Achieved 30% success rate in real-world pruning tasks.

Controller trained in simulation successfully transfers to real orchard environment.

Uses optical flow from wrist-mounted camera for precise pruning guidance.

Abstract

Dormant tree pruning is labor-intensive but essential to maintaining modern highly-productive fruit orchards. In this work we present a closed-loop visuomotor controller for robotic pruning. The controller guides the cutter through a cluttered tree environment to reach a specified cut point and ensures the cutters are perpendicular to the branch. We train the controller using a novel orchard simulation that captures the geometric distribution of branches in a target apple orchard configuration. Unlike traditional methods requiring full 3D reconstruction, our controller uses just optical flow images from a wrist-mounted camera. We deploy our learned policy in simulation and the real-world for an example V-Trellis envy tree with zero-shot transfer, achieving a 30% success rate -- approximately half the performance of an oracle planner.