Abstract: In order to make the path planning of picking robots complex and smooth in extreme temperatures, a 3D path planning method based on 5G communication is proposed. By establishing a three-dimensional spatial environment model under extreme temperatures, the distribution of obstacles and temperature changes under low, normal, and high temperature conditions are simulated to ensure the authenticity and adaptability of the environmental model. By utilizing 5G communication technology, three-dimensional spatial signals and the position information of robot coordinate points can be transmitted to the network in a timely manner. Using the Next Best View （NBV） algorithm based on deep reinforcement learning, the observation angle and movement route of the robot are optimized by utilizing information gain, consumption cost, etc., making the path smoother and more efficient. Through simulation experiments, it is known that this method can generate relatively short paths and improve the smoothness of the paths under extreme temperatures. Applying this method to practical applications, it exhibits excellent path planning performance and robustness in complex environments, providing reliable technical support for the application of picking robots under extreme temperature conditions.