Design, kinematic modeling, and hybrid AI-based optimization of a vision-guided 6-DOF robotic arm for capsicum harvesting in greenhouse environments
Ayan Paul, Rajendra Machavaram
Abstract
This study presents a comprehensive design and optimization framework for a 6-DOF robotic manipulator specifically developed for vision-guided capsicum harvesting in greenhouse environments. The robotic arm was mechanically designed using a hybrid material strategy comprising polylactic acid plus (PLA+) structural links, thermoplastic polyurethane (TPU) gripper fingers, and a stainless-steel scissor cutter, tailored to maneuver within a canopy width of 30–60 cm and plant height of 60–100 cm. The actuation system combines NEMA 17 and NEMA 23 stepper motors with TB6560 drivers and a 180° servo end-effector, supporting a rated payload of 200 g for precise gripping and peduncle cutting. A classical inverse kinematics (IK) solution was derived using the Denavit–Hartenberg (DH) convention for the R–R–R–R–R–R configuration, enabling deterministic mapping from peduncle pose to joint angles θ 1 – θ 6 . Workspace evaluation using a Unified Robot Description Format (URDF) based model in MATLAB/Simulink demonstrated a maximum reach of 0.55 m for the 6-DOF arm, compared to 0.46 m and 0.39 m for reduced 5-DOF and 4-DOF variants, respectively, validating the advantage of full articulation in occlusion-prone canopies. To enhance IK accuracy, a hybrid AI-based optimization pipeline was implemented, incorporating a neural network trained over 100 epochs (MSE ≈ 0.01), fuzzy logic correction (mean error reduced from 3.46 mm to 2.01 mm), and genetic algorithm refinement (85% of solutions within 1.75–2.25 mm). The integrated system enables high-precision, real-time harvesting with minimal mechanical complexity, and reliable performance in structured crop environments.