In practical applications, the boiler wall-climbing robot has some problems, such as limited use of wheel encoders and inertial measurement units (IMU), and easy slippage between the wheels and the metal water wall. In the view of the above factors leading to the decline of robot global localization and pose tracking performance, a robot global localization method based on the laser odometry and the improved adaptive Monte Carlo localization (AMCL) is proposed. Firstly, the laser odometry based on PL-ICP method is adopted to replace the traditional wheel and inertia odometry. Secondly, the idea of DNA cross mutation in genetics is introduced into the particle sampling process of AMCL algorithm to design an improved AMCL method based on genetic algorithm, which is able to alleviate the problems of posture tracking performance degradation and slow location recovery caused by AMCL particle depletion. Experimental results show that the absolute localization error of this method is controlled within 12. 7 cm and the accuracy is 32. 4% higher than that of AMCL method. The localization result of this method is almost unaffected when the robot slips slightly. The speed of the system to restore the localization is 35% higher than that of the ordinary AMCL method when the robot slips greatly.