Abstract: Aiming at the problems of low efficiency and poor environmental adaptability in automatic assembly of industrial robots,a robot assembly strategy based on geometric constraint and the Hidden Markov Model (HMM) was proposed to accomplish peg-in-hole assembly tasks.The theoretical track of peg-in-hole assembly was obtained by establishing geometrical and mechanical modelling under different contact state and analyzing the geometric constraint and stress characteristic in the assembly process.To ensure the accuracy and robustness in actual assembly process,an admittance controller was implemented to trace the required force,which was decided by Learning from Demonstration (LFD) based on Hidden Markov Model and Gaussian Mixture Regression (HMM-GMR) with few samples.The peg-in-hole experiments were conducted aiming at the clearance of the peg and the hole of 0.16 mm,which was verified the effectiveness of the assembly strategy in large positional deviations and few training samples.

Key words: peg-in-hole assembly, geometric constraint, hidden Markov model, learning from demonstration, admittance control

摘要： 针对工业机器人自动装配效率低、环境适应性差等问题,提出一种基于几何约束物理模型与隐马尔可夫链模型(HMM)的机器人轴孔装配策略。首先,通过建立轴孔各接触状态下几何与力学模型,分析零件装配过程中几何约束与受力特点,以获得轴孔装配策略的理论装配轨迹。其次,应用基于隐马尔可夫链模型和高斯混合回归(HMM-GMR)的少样本示教学习方法获得装配过程中的期望接触力,通过导纳控制器对理论轨迹进行补偿,实现对期望接触力的跟踪,从而保证实际装配过程中接触运动的准确性和鲁棒性。最后,针对最小间隙为0.16 mm的轴孔零件进行装配实验,验证了在少训练样本、较大定位偏差情况下所提出装配策略的有效性。

关键词: 轴孔装配, 几何约束, 示教学习, 隐马尔可夫链模型, 导纳控制