Learning-based object center adjustment for 3D reconstruction using an eye-in-hand robotic manipulator

Patino Minan, Jose, Duman, Furkan, Romero Cano, Victor ORCID: https://orcid.org/0000-0003-2910-5116, Lai, Yukun ORCID: https://orcid.org/0000-0002-2094-5680 and Hernandez Vega, Juan ORCID: https://orcid.org/0000-0002-9593-6789 2026. Learning-based object center adjustment for 3D reconstruction using an eye-in-hand robotic manipulator. Presented at: Towards Autonomous Robotics Systems, Manchester, UK, 7–9 September 2026. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only Download (8MB)
	PDF (Provisional file) - Accepted Post-Print Version Download (17kB)

Abstract

Object reconstruction (OR) is a fundamental task for applications like manufacturing, industrial quality control, and cultural preservation. Nowadays, the need to automate this task for repetitive processes has led to the use of robotic manipulators that can move an eye-in-hand camera to reconstruct objects. Even though there are multiple approaches to establish the views that are needed to create 3D models, such approaches are limited by several assumptions such as free movement of the camera, optimal position of the object that makes all views reachable, or the use of additional mechanisms that help the robot to reach such views. In this paper, we present an approach for OR driven by an eye-in-hand fixed-base manipulator, which accounts for non-optimal reachability conditions of the initial views. To do so, our approach leverages a partial reconstruction, which is obtained from the initially reachable views, and determines an object displacement that aims to maximize the total number of reachable views, thus improving the final OR. Our approach uses a machine learning (ML)-based strategy to determine such displacements. We evaluate our approach by reconstructing different objects in simulation and real-world tests, in which the coverage metrics and surface area demonstrate better final reconstructions, compared to the reconstruction system without using the adjustment approach.

Item Type:	Conference or Workshop Item - published (Paper)
Status:	In Press
Schools:	Schools > Computational & Mathematical Sciences Schools > Computer Science & Informatics
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Date of First Compliant Deposit:	14 August 2026
Date of Acceptance:	17 June 2026
Last Modified:	20 Aug 2026 09:27
URI:	https://orca.cardiff.ac.uk/id/eprint/188982

Actions (repository staff only)

Edit Item