<p>To address the lack of comprehensive consideration of real-world scenarios during disassembly-sequence planning and the difficulty of quantifying automation needs in human–robot collaborative disassembly, a knowledge graph based method with improved topological sorting was developed for electric vehicle battery (EVB) disassembly during this study. This method enhances efficiency and safety during disassembly. To define constraints related to the disassembly priority, assembly, and safety, an ontology model was constructed for the battery. Data annotation and deep learning techniques were used for entity recognition and relationship extraction, which were in turn used for the formation of a battery-disassembly knowledge graph (KG). To effectively allocate the various tasks to either humans or robots, an automation-feasibility assessment method was also developed. To further optimize the disassembly efficiency and minimize the subjective bias, the entropy weighting method was used to scientifically distribute the weights of the influencing factors, which included human operation difficulty, disassembly safety, disassembly time, and disassembly cost. Mapping these data to the knowledge graph enabled dynamic updates. Based on this foundation, an improved topological sorting method that utilizes the Tarjan algorithm was employed to generate a rational disassembly sequence in which the removal of components followed all the dependency constraints. The proposed approach was first validated using the Audi A3 EVB case and then compared with a similar algorithm applied to a different EVB. The results showed that the proposed method produced an optimal disassembly sequence about 1.0% longer than the reference method due to consideration of risk and human factors. However, when human–robot collaboration was applied, the total time was reduced by 7.22%, demonstrating enhanced efficiency and safety.</p>