<p>Visual servoing in dual-arm robotic systems remains challenging in unstructured environments due to dynamic uncertainties, limited perception accuracy, and inefficient arm coordination. To address these limitations, this paper proposes a neuro-inspired, priority-based dual-arm visual servoing framework for the Baxter robot, integrating deep perception and adaptive control to achieve robust real-time target tracking in unstructured environments. The system employs a lightweight YOLOv4-tiny network for high-speed object detection, trained on a custom-labeled dataset using MakeSense, and leverages Kinect v2 for precise 3D localization via depth sensing. A refined hand-eye calibration method, utilizing an ’eye-outside-the-hand’ configuration, enables accurate spatial transformation between camera and robot coordinate frames. To ensure stability in dynamic tracking scenarios, a Kalman filter is applied to mitigate noise and enhance prediction accuracy. A novel dual-arm coordination mechanism is introduced, incorporating a priority-based task allocation strategy with smooth transition control and real-time arm switching based on workspace partitioning and confidence-weighted heuristics. Furthermore, a Radial Basis Function (RBF) neural network-based Function Approximation Technique (FAT) is integrated into the control loop to compensate for nonlinearities and unknown dynamics, enabling adaptive motion regulation. Extensive experiments validate the proposed system’s effectiveness, demonstrating high-precision, low-latency visual servoing across 3D space, as well as stable and efficient dual-arm coordination under varying target trajectories.</p>