Overview

A complete simulation pipeline for autonomous fruit harvesting using a UR10 manipulator in Gazebo.

The system integrates perception, motion planning, and task orchestration into a deterministic workflow.


System Architecture

  • UR10 robot model with custom orchard environment
  • YOLO-based apple detection
  • 3D pose estimation from RGB camera stream
  • MoveIt 2 motion planning with collision avoidance
  • State machine driven task planning
  • Pick → place execution pipeline

Engineering Focus

  • Clean separation between perception and control layers
  • Robust collision checking within cluttered tree environment
  • Deterministic task sequencing
  • Simulation-first validation for deployment readiness

Outcome

A full-stack manipulation system demonstrating perception-to-action autonomy in a constrained agricultural scenario.

Building a perception-driven manipulation pipeline that needs to work without manual tuning per object?

Getting detection, pose estimation, motion planning, and task sequencing to work together reliably in a cluttered environment is harder than any single piece in isolation. This project was about making that pipeline deterministic — from camera feed to confident pick execution — without hand-tuning for each new scenario.

If you are building a manipulation system that needs to go from perception to reliable action in an unstructured environment, we can help.

Tell us about your project

Or email kodorobotics@gmail.com