摘要
工业生产中装配工艺直接关系到产品的产能和质量。当前机器人化装配算法需要根据具体的装配任务进行示教,难以适应工业场景下迭代迅速、工艺多变等特点。针对上述难题,提出一种基于强化学习的机器人装配算法。首先,使用机器人末端执行器力-力矩传感器和视觉传感器的多模态数据,提升模型的感知能力。然后,针对机器人试错成本高的问题,提出了一种融合人工示例先验的强化学习训练算法,通过专家经验池对策略模型和价值模型的参数进行初始化以减少低效探索。最后,在机器人齿轮装配任务中对所提算法进行验证。实验结果表明,加入多模态感知数据的模型具有更强的鲁棒性,人工示例先验的融合能够显著提升算法的训练效率。
Assembly process is directly related to the productivity and quality of products.Current robotic assembly algorithm needs to be taught according to specific assembly tasks,and it is difficult to adapt to the characteristics of rapid iteration and variable process in industrial scenarios.To solve these problems,an industrial robot assembly algorithm based on reinforcement learning(RL)is proposed.Firstly,multi-modal data from force-torque sensors of the robot end-effectors and vision sensors are used to improve the model perception capabilities.Then,for the high cost of robot trial and error,a RL algorithm integrating artificial demonstration priors is proposed,and the parameters of strategy model and value model are initialized through the expert experience pool to reduce inefficient exploration.Finally,the proposed algorithm is validated on a robotic gear assembly task.The experimental results show that the model with multi-modal perception data has stronger robustness,and the fusion of artificial demonstration priors can significantly improve the training efficiency of the algorithm.
作者
刘行
黄庭安
董云龙
沈檀
LIU Xing;HUANG Ting’an;DONG Yunlong;SHEN Tan(School of Artificial Intelligence and Automation,Huazhong University of Science and Technology,Wuhan 430074;Center for Intelligent and Networked Systems,Tsinghua University,Beijing 100084)
出处
《控制工程》
CSCD
北大核心
2023年第7期1308-1316,共9页
Control Engineering of China
基金
国家重点研发计划资助项目(2018YFB1701202)
关键词
智能装配
强化学习
示例先验
机器人
Intelligent assembly
reinforcement learning
demonstration priors
robotics
