Design of an Autonomous Drug Delivery Robot Based on YOLO-v8

To meet the demands for drug delivery efficiency and safety in medical settings, this study designs an autonomous drug delivery robot system based on YOLO-v8. The YOLO-v8 algorithm is applied to drug bottle multi-pose recognition and robot collaborative control scenarios. The system utilizes a heterogeneous hardware architecture, integrating the YDLIDAR 4ROS laser radar and Nuwa-HP60C depth camera on a Jetson Nano host for environmental perception. The STM32F407 microcontroller drives a mecanum wheel chassis and a six-axis robotic arm to perform robot movement and drug picking tasks. To improve the robustness of drug recognition, a dynamic lightingangle coupling dataset construction method is proposed. This involves collecting $\mathbf{9, 7 2 0}$ real drug bottle images under five natural lighting conditions (color temperature $\mathbf{2, 7 0 0 - 6, 5 0 0 K}$) and a $\boldsymbol{\pm} \mathbf{4 0}{ }^{\circ}$ dual-axis rotation range, covering 20 different drug types. Experiments show that YOLO-v8 achieves a $94.19 \%$ average recognition accuracy in this scenario, an improvement of $1.87 \%$ over YOLO-v5. Integrated system tests demonstrate that the robot performs 50 trials in a $7 \mathrm{~m} \times 3 \mathrm{~m}$ simulated ward environment with a drug picking success rate of $94 \%$. This work validates the engineering applicability of YOLO-v8 in medical logistics, offering a scalable technological solution for hospital intelligent transformation.