Orchard fruit-pick teleoperation capture

This is a custom capture program for teleoperated fruit picking in commercial orchards. Human operators demonstrate reach, grasp, and detach across real canopy occlusion, cluster density, and lighting while we record synchronized RGB-D and gripper state, tagged with detach outcomes. It exists because simulation rarely reproduces leaf cover, fruit compliance, and outdoor illumination, which is where harvest policies fail.

What we collect

Human-guided picks with synchronized depth and gripper state across varieties, ripeness, and row positions. Episodes include clean detaches alongside slips, crushed fruit, and occluded approaches so evaluation reflects harvest success, not a sanitized subset.

How capture works

A pilot week validates camera placement under canopy and the outcome taxonomy, then capture scales across rows and lighting windows. The workflow follows our robotics data collection process: scope, pilot, and scale, with teleoperation capture as the demonstration method.

Who it is for

Agricultural robotics teams training harvest policies where simulation misses real canopy variance. This scenario belongs to agricultural robotics data collection and pairs with crop-row weeding navigation when the same platform must both navigate rows and pick.