摘要
采用智能机器人完成无序抓取任务(Bin-picking)已成为工业自动化的上下料场景的新型解决方案,而从散乱零件中识别目标物体并完成姿态估计是解决此类问题的关键。为解决高度自动化的工业上下料场景中的机器人无序抓取任务,借助基于物体渲染的样本合成方法,提出用于姿态估计样本的改进CycleGAN迁移模型,增加了基于物体掩码的内容一致性损失,实现了无需真实域标注和无配对样本下的姿态估计样本的跨域迁移,并提出了基于姿态迁移学习的机器人无序抓取方法,进行了无需人工标注的物体6D姿态估计,实现了高度自动化的模型训练流程和简单易用的模型部署流程。同时,该文方法经实际场景下的抓取实验证明其切实有效。
It has become one of the research hotspots in industrial automation to perform Bin-picking for machine tending with industrial robots and intelligent sensor.Accurate detection and object pose estimation from clutter scene is one of the key problems to solve Bin-picking task for industrial robot.To handle the robot Bin-picking problem in automated industrial machine tending,based on the synthetic samples from rendering,this paper proposes an improved CycleGAN model for pose estimation samples,which preserves structural information better and realizes the cross-domain conversion of pose estimation samples.Then,this paper proposes a robot Bin-picking pipeline based on transfer learning for object pose,which realizes a highly automated model training pipeline and a practical pipeline.
出处
《工业控制计算机》
2022年第11期93-95,共3页
Industrial Control Computer
基金
江苏省自然科学基金(BK20201264)。
