%0 Generic %1 chen2022simtoreal %A Chen, Kai %A Cao, Rui %A James, Stephen %A Li, Yichuan %A Liu, Yun-Hui %A Abbeel, Pieter %A Dou, Qi %D 2022 %K pose-estimation sim2real %T Sim-to-Real 6D Object Pose Estimation via Iterative Self-training for Robotic Bin-picking %U http://arxiv.org/abs/2204.07049 %X In this paper, we propose an iterative self-training framework for sim-to-real 6D object pose estimation to facilitate cost-effective robotic grasping. Given a bin-picking scenario, we establish a photo-realistic simulator to synthesize abundant virtual data, and use this to train an initial pose estimation network. This network then takes the role of a teacher model, which generates pose predictions for unlabeled real data. With these predictions, we further design a comprehensive adaptive selection scheme to distinguish reliable results, and leverage them as pseudo labels to update a student model for pose estimation on real data. To continuously improve the quality of pseudo labels, we iterate the above steps by taking the trained student model as a new teacher and re-label real data using the refined teacher model. We evaluate our method on a public benchmark and our newly-released dataset, achieving an ADD(-S) improvement of 11.49% and 22.62% respectively. Our method is also able to improve robotic bin-picking success by 19.54%, demonstrating the potential of iterative sim-to-real solutions for robotic applications.

@misc{chen2022simtoreal, abstract = {In this paper, we propose an iterative self-training framework for sim-to-real 6D object pose estimation to facilitate cost-effective robotic grasping. Given a bin-picking scenario, we establish a photo-realistic simulator to synthesize abundant virtual data, and use this to train an initial pose estimation network. This network then takes the role of a teacher model, which generates pose predictions for unlabeled real data. With these predictions, we further design a comprehensive adaptive selection scheme to distinguish reliable results, and leverage them as pseudo labels to update a student model for pose estimation on real data. To continuously improve the quality of pseudo labels, we iterate the above steps by taking the trained student model as a new teacher and re-label real data using the refined teacher model. We evaluate our method on a public benchmark and our newly-released dataset, achieving an ADD(-S) improvement of 11.49% and 22.62% respectively. Our method is also able to improve robotic bin-picking success by 19.54%, demonstrating the potential of iterative sim-to-real solutions for robotic applications.}, added-at = {2022-07-17T17:07:51.000+0200}, author = {Chen, Kai and Cao, Rui and James, Stephen and Li, Yichuan and Liu, Yun-Hui and Abbeel, Pieter and Dou, Qi}, biburl = {https://www.bibsonomy.org/bibtex/22c9696416c2ef42a2b5f58c121390a56/redtedtezza}, description = {Sim-to-Real 6D Object Pose Estimation via Iterative Self-training for Robotic Bin-picking}, interhash = {617e41d4dddc5f538d95fdbb2b66ed00}, intrahash = {2c9696416c2ef42a2b5f58c121390a56}, keywords = {pose-estimation sim2real}, note = {cite arxiv:2204.07049Comment: 25 pages, 16 figures, Project homepage: www.cse.cuhk.edu.hk/~kaichen/sim2real_pose.html}, timestamp = {2022-07-17T17:07:51.000+0200}, title = {Sim-to-Real 6D Object Pose Estimation via Iterative Self-training for Robotic Bin-picking}, url = {http://arxiv.org/abs/2204.07049}, year = 2022 }