In recent years, vision-based gesture adaptation has attracted great attention from many experts in the field of human-robot interaction, and many methods have been proposed and successfully applied, such as particle swarm optimization and genetic algorithm. However, the reduction of the error and energy consumption of a robot while paying attention to more subtle attitude changes is very important and challenging. In view of these problems, we propose a population randomization-based multi-objective genetic algorithm. The gesture signal is processed with a slight change by imitating the biological evolution mechanisms. In the proposed algorithm, a random out-of-order matrix is added in the process of population evolution synthesis to prevent the premature grouping convergence of the new population. The weights of the objective function and the elite retention strategy are adopted, and the most adaptable individuals in each generation are inherited directly in the next generation without any recombination or mutation. To verify the effectiveness of the algorithm, preliminary application experiments are performed on the gesture adaptation of a robotic arm. The results are compared with the original signal, and the comparison shows that by using the proposed method, the energy consumption is reduced, and the end error is decreased to less than 3 mm while ensuring the tracking effect of the robotic arm. These obtained results meet the communication requirements for human-robot interactions such as handshakes. Moreover, the proposed method has better performance, uses less energy, and has a smaller tracking error than the particle swarm optimization, the single-objective genetic algorithm, and the traditional multi-objective genetic algorithm. A preliminary application experiment indicates that the robotic arm can adapt to human gestures in real time.

近年来，基于视觉的手势自适应已经引起了人机交互领域众多专家的广泛关注，并且提出并成功地应用了诸如粒子群优化和遗传算法等多种方法。然而，在关注更细微的姿态变化的同时减少机器人的误差和能量消耗是非常重要且具有挑战性的。针对这些问题，我们提出了一种基于种群随机化的多目标遗传算法。通过模仿生物进化机制对手势信号进行轻微的处理。该算法在种群进化合成过程中增加了随机无序矩阵，以防止新种群的过早分组收敛。采用目标函数和精英保留策略的权重，并且每一代中最适应的个体都可以直接在下一代中继承，而无需任何重组或突变。为了验证该算法的有效性，对机器人手臂的手势适应进行了初步的应用实验。将结果与原始信号进行比较，比较结果表明，所提方法降低了能耗，并且在保证机器人手臂跟踪效果的同时，将终端误差减小到3mm以内。这些获得的结果满足了诸如握手之类的人机交互的通信要求。此外，所提出的方法具有更好的性能，使用更少的能量，与粒子群优化，单目标遗传算法和传统的多目标遗传算法相比，跟踪误差更小。初步的应用实验表明，机械臂可以实时适应人类手势。