Objective: In order to deal with the impact of cyber attack on the cyber-physical system of food picking robot. Methods: A method for states estimation and attacks detection of food picking robot system based on interval observer was proposed. Firstly, the digital model of the food picking robot located in the cyber-physical system was established, using Lagrange method to establish the dynamic equation of the manipulator and convert it into the state-space model, the network attacks model was constructed according to the feedback control strategy. Considering the nonlinear characteristics and noiseand applying the generalized system theory was applied to transform sensor attack into the system state, and the interval observer was designed as the attack detector, and the stability of errors system was proved. Results: The simulation results showed that observer output was close to the given real signal. The proposed method was correct and effective. The sensor attack can be reconstructed and the actuator attack can be detected from interval estimation results. Conclusion: The proposed method can realize states estimation and actuator attacks detection of food picking robot in cyber-physical system environment.