From Poking to Grasping: Active Shape Completion for Improved Robot Manipulation

Lukas Rustler, Matej Hoffmann

Keywords: shape completion, manipulation, active perception, grasping

TL;DR: Active shape completion for improved robot manipulation.

Abstract: Robust robot grasping of a priori unknown objects remains challenging when the object is only partially observed. While single-view shape completion methods infer missing geometry, their predictions often contain high uncertainty that leads to grasp failures. Active perception solves this by gathering new targeted information. This extended abstract synthesizes progress using a visuo-tactile framework across three progressive methodologies---Act-VH, VISHAC, and ShapeGrasp. We outline the transition from uncertainty-driven haptic poking to a closed-loop "grasp-and-complete" pipeline, demonstrating how physical interaction and grasp attempts can seamlessly serve as active perception queries to refine implicit shape models and achieve state-of-the-art manipulation success.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 3