Safe, Occlusion-Aware Manipulation for Online Object Reconstruction in Confined Spaces

Method

Our method defines an Occlusion Dependency Graph (ODG) data structure to represent the occlusion relationship among objects. By following the topological order of ODG, it is able to identify safe objects to reconstruct first and gradually reconstruct all objects in the scene. To handle reachability and visibility issues, our method rearranges objects that block or occlude the target object, which is essential to safely extract the target while reducing occlusion in the scene.

Results

 •ODG highlights objects that are being moved as Green, and objects that have been reconstructed as Red.

•Bottom-left two figures show the reconstructed object model, one using optimistic volume (object surface), and one using conservative volume (object occupied space + unseen parts). TSDF is used for object reconstruction. The bottom-right figure shows the change of scene occlusion as objects are removed.

•Several waypoints are sampled for sensing for object reconstruction.

•For clearer videos, please see the video link below.

Source Code & Supplementary Material

1. Paper: https://arxiv.org/abs/2205.11719

2. GitHub: https://github.com/MiaoDragon/pracsys-vbcpm

3. Video: https://youtu.be/Wv7dJ3AcsrM