Visuomotor Robotic Pruning in Planar Orchards Using Hybrid Reinforcement Learning

Tue, 01 Sep 2026 00:00:00 +0000

Watch the Demo

See the visuomotor policy achieving zero-shot sim-to-real transfer in real commercial and experimental orchards:

Abstract

Dormant tree pruning is labor-intensive yet essential for maintaining modern high-productivity fruit orchards. In this work, we focus on pruning of modern planar tree training systems—V-Trellis apples and UFO cherries—where trunks and primary branches are trained into approximately planar walls. We introduce an end-to-end pipeline to learn a closed-loop visuomotor controller for robotic pruning, trained entirely in simulation and deployed in real orchards zero-shot.

Key contributions:

Novel orchard simulator: Physics-based PyBullet environment with synthetically generated planar orchard tree meshes (Tree-L-Py)
Automated demonstration generation: Motion planning pipeline (RRT-Connect) to collect successful pruning trajectories as offline data
Hybrid-PPO: A novel hybrid reinforcement-learning algorithm combining offline expert demonstrations with online PPO rollouts
Optical-flow perception: Avoids 3D reconstruction; wrist-mounted camera optical flow enables strong sim-to-real generalization
Real-world validation: 38 physical trials (28 outdoor field trials in commercial orchards + 10 indoor lab tests) with zero-shot sim-to-real transfer; outperforms RRT-Connect baseline on hardware

Simulation results: 49.9% success on V-Trellis apples and 46.0% on UFO cherries (3,000 cutpoints). Real-world results: 71% success on V-Trellis apples in outdoor commercial orchards.

Deep Reinforcement Learning |

Visuomotor Robotic Pruning in Planar Orchards Using Hybrid Reinforcement Learning

Watch the Demo

Abstract