Robotic automated disassembly is a key process in the current remanufacturing industry. However, in complex scenarios, it is challenging to establish accurate control models for robot disassembly based on mechanism models. Therefore, this paper proposes a novel robotic automated disassembly method using reinforcement learning. By utilizing a reinforcement learning model, the robot's action selection in the disassembly process is optimized in a data-driven manner, eliminating the need to explicitly define the mechanical relationships of the disassembly process. Using peg hole disassembly as an example, a reinforcement learning training and application model for peg-hole disassembly is established based on the disassembly workflow and reinforcement learning principles. This model is deployed in a physical environment for training and testing. Experimental results demonstrate the effectiveness of the proposed method in the peg-hole disassembly task.