Learning to Prune Branches in Modern Tree-Fruit Orchards

2025-07-30

AI Summary

Pruning high-yield orchards during dormancy is labor-intensive, and disordered branch arrangements hinder robotic localization and pose control. Method: This paper proposes a closed-loop pruning control system leveraging optical flow visual feedback. Departing from conventional 3D reconstruction–dependent paradigms, it directly employs wrist-mounted camera optical flow as input and trains a visuomotor controller via deep reinforcement learning in a simulation environment geometrically faithful to real orchards, enabling zero-shot transfer to actual V-Trellis Envy apple trees. Contribution/Results: The core innovation lies in real-time branch orientation estimation and precise end-effector positioning at cut points—while maintaining cutter-axis orthogonality to target branches—using only sparse optical flow. Experiments demonstrate a 30% pruning success rate in real-world settings, approximately 50% of an ideal planner’s performance, thereby establishing, for the first time, the feasibility of end-to-end, 3D-reconstruction–free, optical-flow–driven robotic pruning in orchards.

Technology Category

• Intelligent Robots: Manipulation
• Computer Vision: Vision for Robotics & Autonomous Driving
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

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.

Problem

• Develop robotic pruning for labor-intensive orchard maintenance
• Guide cutter accurately in cluttered tree environments
• Use optical flow images instead of 3D reconstruction

Innovation

• Closed-loop visuomotor controller for pruning
• Training with novel orchard simulation
• Uses optical flow images, not 3D reconstruction