Design and Laboratory Evaluation of a Lightweight Long-Reach Manipulator for Vision-Guided Positioning in Rabbit Feeding

Automated feeding in cage-based rabbit production is challenging because cage geometry and limited workspace require a compact manipulator with sufficient reach, low mass, and reliable task-level positioning. This study developed a lightweight five-degree-of-freedom manipulator integrated with a vision-guided positioning framework for cage-based rabbit feeding applications. The prototype achieved an effective task reach of 1.20 m with a total mass of 12.37 kg. The design incorporated lightweight carbon-fiber-reinforced polymer (CFRP) long-span links, integrated actuators selected according to joint torque requirements, task-constrained kinematics, closed-form inverse kinematics, and Cartesian trajectory planning. A red–green–blue and depth (RGB-D) localization interface used four ordered feed-port keypoints and a horizontal-plane constraint to provide three-dimensional upper-tier task targets for the manipulator. All representative task points were numerically reachable under the defined software joint limits, and the maximum planned joint velocity was 0.525 rad s−1. Component-level linear-static finite-element analysis predicted no yielding in the four analyzed aluminum components under the simplified conservative 5 kg verification load case. Static loading of the assembled arm produced mean downward displacements of 3.0–20.0 mm at the J5 flange under 1–5 kg loads. Laboratory evaluation demonstrated three-dimensional repeatability radii of 1.68–4.63 mm. Across four upper-tier locations, the mean Euclidean feed-port localization error ranged from 9.01 to 20.47 mm, with all 80 trials satisfying the ±20 mm-per-axis criterion. The mean tool center point (TCP)-equivalent end-point error ranged from 18.23 to 37.15 mm. These laboratory results support the feasibility of the developed manipulator for upper-tier vision-guided positioning in rabbit-feeding geometry, while indicating that multi-point calibration and spatial compensation are needed to further reduce system-level positioning error. Validation was limited to laboratory testing; no feed was discharged and no rabbit-house operation was evaluated.