Self-Supervised Unseen Object Instance Segmentation via Long-Term Robot Interaction
This work addresses the challenge of segmenting unseen objects in cluttered environments for robotics applications, representing an incremental advancement over prior single-action methods.
The paper tackles the problem of unseen object instance segmentation in real-world robotics by using long-term robot pushing interactions to collect data, which improves segmentation accuracy for objects in close proximity. The result is a significant improvement in segmentation accuracy both within and across domains, and enhanced robotic grasping performance for unseen objects.
We introduce a novel robotic system for improving unseen object instance segmentation in the real world by leveraging long-term robot interaction with objects. Previous approaches either grasp or push an object and then obtain the segmentation mask of the grasped or pushed object after one action. Instead, our system defers the decision on segmenting objects after a sequence of robot pushing actions. By applying multi-object tracking and video object segmentation on the images collected via robot pushing, our system can generate segmentation masks of all the objects in these images in a self-supervised way. These include images where objects are very close to each other, and segmentation errors usually occur on these images for existing object segmentation networks. We demonstrate the usefulness of our system by fine-tuning segmentation networks trained on synthetic data with real-world data collected by our system. We show that, after fine-tuning, the segmentation accuracy of the networks is significantly improved both in the same domain and across different domains. In addition, we verify that the fine-tuned networks improve top-down robotic grasping of unseen objects in the real world.