A power emergency robot can replace the human to enter a dangerous scene to carry out emergency operations. The rationality of robot path planning has great influence on the timeliness and accuracy of its operation. This paper presents a path planning method for an emergency robot which combines an Analytic Hierarchy Process (AHP), improved A algorithm and gradient descent method. First, the operation target in the complex scene is determined by the AHP, and then combined with a hazards risk curve function and robot protective threshold, the cost function of the A algorithm is improved to realize risk-oriented global path planning. Then, based on the gradient descent method, the local dynamic path planning in the process of movement is completed. Simulation results show that this method can choose the operation target independently, avoid operational risk and reach the target quickly, and effectively improve the timeliness and safety of power emergency operation. This work is supported by the National Natural Science Foundation of China (No. 51507157).