Omni-Scan

Creating Visually-Accurate Digital Twin Object Models
Using a Bimanual Robot with Handoff and Gaussian Splat Merging

Tianshuang Qiu*
Zehan Ma*
Karim El-Refai*
Hiya Shah
Justin Kerr
Chung Min Kim
Ken Goldberg

UC Berkeley

* Denotes Equal Contribution

IROS 2025

paper
Paper
code
</Code>
data
Data

TL;DR: Omni-Scan uses a bimanual robot to scan objects from multiple viewpoints, then hands off the object to the other arm getting views of previously occluded points. It then merges the scans into a single, visually-accurate 3DGS model.

Interactive Omni-Scans

Click and move me!

Unmerged and Merged Scans

Click and move me!

Left Scan Remote

Click and move me!

Right Scan Remote

Click and move me!

Merged Remote

Visual Part Inspection

Click and move me!

PVC Pipe Connector A

Click and move me!

PVC Pipe Connector B

Click and move me!

PVC Pipe Connector C

Geometric Part Inspection

Click and move me!

Wireplug A

Click and move me!

Wireplug B

Click and move me!

Wireplug C

Click and move me!

Flashlight A

Click and move me!

Flashlight B

Click and move me!

Flashlight C