ICRA - Generative Digital Twins

About
This workshop will explore generative digital twin technologies as a pathway to overcoming the persistent gaps between reality and simulation (Real2Sim) and simulation and reality (Sim2Real) in robotics. Despite advances in transfer learning and physics-based simulation, robot policies often underperform when deployed in the real world due to unmodeled dynamics, sensory mismatches, and environmental complexity. Digital twins - high-fidelity, adaptive virtual representations of robots and environments - offer a powerful approach to bridging these divides.

We will focus on generative digital twins, where AI-driven models automatically construct, update, and diversify twins to enhance robustness, adaptability, and scalability. Topics will include: (i) pipelines for digital twin generation (Real2Sim); (ii) benchmarks and case studies for Sim2Real transfer; (iii) learning algorithms leveraging adaptive twins; and (iv) applications across industrial, household, and mobile robotics. The workshop will bring together leading researchers from robotics, machine learning, and computer graphics, fostering interdisciplinary exchange through invited talks, poster sessions, and panel discussions. By highlighting recent advances and charting future directions, this event aims to define a roadmap for how generative digital twins can enable more reliable and generalizable robotic intelligence: Digital Twin Generation from Visual Data: A Survey

Schedule June 1st, 2026 - Vienna, Austria

Time		Info
9:00		Organizers: Introductory Remarks
9:30		Keynote: Ingmar Posner Oxford Title: blank Abstract blank
10:00		Keynote: Oier Mees Microsoft Title: blank Abstract blank
10:30		Coffee break
11:00		Keynote: Yuke Zhu NVIDIA, University of Texas at Austin Title: blank Abstract blank
11:30		Keynote: Jiajun Wu Stanford Title: Physical Grounding of Generative Digital Twins for Robotics Abstract Much of our visual world has its intrinsic, physical structure: scenes are made of objects; objects have their geometry, texture, material, and physical properties. With the rapid development of neural visual generative models, what role does such structural information play, or do we still need it at all? In this talk, I will discuss our recent efforts to build generative digital twins that are not only visually realistic, but also physically grounded. Beyond reconstruction and generation, I will also demonstrate how these objects can bridge the sim-to-real gap for robotic manipulation.
12:00		Spotlight talks
12:30		Lunch & Poster session
14:00		Keynote: Manolis Savva SFU Title: blank Abstract blank
14:30		Keynote: Hengshuang Zhao The University of Hong Kong Title: blank Abstract blank
15:00		Keynote: Ken Goldberg UC Berkeley Title: blank Abstract blank
15:30		Coffee Break
16:00		Discussion. Panelists: Ingmar Posner, Oier Mees, Yuke Zhu, Jiajun Wu, Manolis Savva, Hengshuang Zhao, Ken Goldberg
17:00		Organizers: Closing Remarks

Call for Papers

Targeted Topics

The gDT-IV Workshop at ICRA 2026 welcomes submissions on all aspects of generating digital twins from images and videos, especially (but not limited to):

Applications of digital twins in robotics, media content creation, AI-driven video synthesis, interior design, construction monitoring, and gaming.
Streaming and compression techniques for sharing large 3D scenes over the web.
Synchronization of the Digital Twin model with the video stream.
3D Gaussian Splatting (3DGS) and its extensions for dynamic scenes.
Lighting variations, shadows, material properties, and reflections in 3DGS reconstruction.
Modeling articulated object parts and their motion constraints from video (e.g., cabinets).
Inpainting and completion techniques for occluded regions in complex indoor scenes.
Retrieval-based and generative hybrid pipelines for 3D object and scene creation.
Learning object properties (mass, softness, friction) from videos.
Physics-informed 3D generation, blending physical simulation with visual generation.
Semantic 3D segmentation for scene understanding (object-level vs. fine-grained).

Submission Information

We welcome submissions through two tracks, both submitted via the same OpenReview portal:

Archival Track: Papers should present original, unpublished work and follow a strict double-blind review. Manuscripts must be prepared with the ICRA 2026 Author Kit and be 4–8 pages (references and appendices may extend beyond 8 pages). Accepted papers will appear in the official ICRA 2026 Workshop Proceedings, and authors will present in person. We follow the same policy as the main ICRA conference for this track. By submitting to this track, the authors acknowledge that it has not been previously published or accepted for publication in substantially similar form in any peer-reviewed venue including journal, conference or workshop, or archival forum. Furthermore, no publication substantially similar in content (defined as having 20 percent or more overlap) has been or will be registered or submitted to this or another conference, workshop, or journal during the review period. Violation of any of these conditions will lead to rejection, and will be reported to the other venue to which the submission was sent. For more details policy, please check the official policy at ICRA 2026.
Non-Archival Track: This flexible track welcomes a wider range of contributions including extended abstracts & short papers (up to 4 pages, suitable for work in progress, negative results, or position papers; references and appendices may extend beyond 4 pages) and previously published work (including papers accepted at ICRA 2026 or other venues in the last year). All submissions should be formatted with the ICRA 2026 Author Kit. Submissions remain double-blind for review but will not appear in proceedings. Thus, papers that have already been published at major conferences are also welcome. Authors may also submit their work to future conferences or journals after acceptance to this workshop.

Submit via OpenReview Real2Sim2Real Submission Portal

Submission Instructions

Authors planning to submit are strongly encouraged to create their OpenReview profiles well in advance of the submission deadline.

Please note OpenReview's moderation policy for newly created accounts:

Profiles created without an institutional email address will undergo a moderation process that may take up to two weeks before activation.
Profiles created with an institutional email address are typically activated automatically.

To avoid last-minute issues with submission access, we recommend registering early and, when possible, using an institutional email address.

Important Dates

Archival Submission Deadline: March 15, 2026 (AOE)
Archival Author Notifications: April 10, 2026 (AOE)
Archival Camera-Ready: April 15, 2026 (AOE)

Non-Archival Submission Deadline: March 15, 2026 (AOE)
Non-Archival Author Notifications: April 10, 2026 (AOE)
Non-Archival Camera-Ready: April 15, 2026 (AOE)