<p>In human–robot interactions, due to the frequent cooperation between robots and human operators, there is a high risk of collision during high-speed operations, which may cause harm to both robots and human operators. Consequently, this paper focuses on the safety control issues of velocity-constrained robotic systems with output measurements and uncertainties. An observer-based safety control strategy, which combines the neural network (NN) with the control barrier function (CBF), is proposed to enhance the reliability and adaptability of robotic safety. To deal with the velocity constraints, a NN-based CBF (NNCBF) is formulated by the state observer and the NN with respect to the robotic system. It is proven that the proposed NNCBF guarantees the forward invariance of the safe set for uncertain robotic systems where only output measurements are available. Subsequently, the safety control scheme is developed by combining the task-based nominal controller with the barrier conditions of the proposed observer-based NNCBF in a quadratic program (QP) framework. The safety controller can be then obtained by solving the QP problem, which is capable of not only ensuring safety but also maintaining the control performance of the robotic system. Finally, the effectiveness of the proposed method is validated through both simulations and experiments.</p>