Robot helps to increase automation and economic benefits of disassembly line systems, and reduce risk to the human worker. For the robotic disassembly line, its energy consumption can be further optimized to reduce carbon dioxide emissions. In this paper, energy consumption of disassembly line systems is considered to be one of optimization objectives of disassembly line balancing problem. In the proposed model, the optimization objectives are to minimize the energy consumption and the line length (number of multi-robotic workstations and number of opened disassembly robots). To solve this multi-objective optimization problem, an improved NSGA-III optimization algorithm which consists of problem-dependent global and local variation operators is proposed. Several experiments are conducted to verify the effectiveness of the proposed method. In terms of hypervolume indicator, compared with three other state-of-art multi-objective evolutionary algorithms, the proposed method outperforms the best in small-scale, medium-scale, and large-scale problems. The proposed method also performs better on the problem of all scales than MOEA\D and NSGA-II in inverted generational distance metric, the proposed approach outperforms NSGA-III in most small-scale, some medium-scale and large-scale problems. The Friedman test based on the indicators of hypervolume and inverted generational distance is also conducted to verify the effectiveness of the proposed method.