With the development of artificial intelligence technology, the application prospect of Embodied AI robots has been widely recognized by the industry and has a vital role in industrial production, scientific research, and other scenarios. Usually, the perception of the environment by an Embodied AI robot is based on the hand-eye calibration system; however, in practice, the vision system is often affected by the attitude of the robot's end-effector and the limitation of the gripper's range of motion, to deal with this problem, we propose a dual-mechanical-arm collaborative grasping system for an intelligent robot, which consists of a six-degree-of-freedom grasping arm and a six-degree-of-freedom searching arm. We construct a kinematic model for the grasping arm based on the DH method, and at the same time, we configure the Fast-RCNN-based part visual recognition algorithm in the search arm to realize the localization of the position of the end apparatus of the grasping arm and construct a smooth grasping arm motion trajectory scheme. We validated the proposed kinematic planning and image recognition methods in both the Matlab and PyTorch frameworks. Additionally, we conducted practical experiments on cooperative grasping with dual arms. The results indicate that the dual-arm cooperative system can achieve high-precision localization of target objects and execute smooth joint motion trajectories for grasping under specific orientations. This effectively enhances the operational accuracy and capability of the robot in grasping scenarios.