FROM PERCEPTION TO ACTION:
REPRESENTATION-CENTRIC ROBOT AUTONOMY

📍 Sydney, Australia
13 July 2026

Overview

This workshop focuses on the central role of representations in shaping the performance, scalability, and reliability of autonomous robotic systems. While perception, mapping, planning, and decision making are often studied as separate components, the choice of scene and map representations fundamentally couples these modules and directly influences downstream robot behavior. This workshop aims to bring together researchers who study how representation design: geometric, probabilistic, dense, implicit, or learned, affects the entire autonomy pipeline, from sensing and localization to planning, navigation, manipulation, and exploration.

The workshop targets an audience spanning multiple robotics sub-communities, including SLAM and mapping, robot perception, motion planning, decision making, manipulation, and autonomous navigation. We particularly encourage participation from researchers working on dense and implicit map representations (e.g., occupancy, signed distance fields, Gaussian splatting, neural fields), task-aware perception and mapping, active and informative planning, and joint optimization of perception and decision processes. Presenters and panelists will be drawn from both geometry- and learning-oriented communities, as well as from researchers bridging classical model-based methods with modern data-driven approaches.

The relevance of this workshop lies in providing a focused forum to examine how different map representations impact core robotic tasks. As robotic systems are deployed in increasingly diverse environments, practitioners are faced with a wide range of representation choices, each offering distinct advantages and limitations for downstream tasks. By centering the discussion on representation design and its practical implications, this workshop aims to compare how different representations support or constrain localization accuracy, planning efficiency, decision robustness, and exploration strategies. Rather than advocating a single representation paradigm, the workshop seeks to clarify trade-offs, failure modes, and task-dependent requirements, and to identify open problems in aligning representations with the needs of closed-loop autonomy. We expect the discussions to help researchers and practitioners make more informed representation choices and to inspire future work on task-aware mapping and perception for autonomous robotic systems.

Call for Contributions

We invite researchers and practitioners from robotics, computer vision, and related fields to share advances in representation-centric autonomy: preliminary results, systems insights, and lessons from deployment. Accepted contributions can be featured in the poster session; accepted papers may be listed on this workshop site.

Submission formats

Formats follow the official RSS 2026 venue.

Research Papers (up to 6 pages, excl. refs)

Field Reports / Visionary Papers (2–4 pages, excl. refs)

Non-anonymous Authors / Single-Bind Review

Submission via OpenReview

Submit through the RSS 2026 workshop group on OpenReview. Use your OpenReview account to create or upload submissions and track decisions.

OpenReview workshop venue →

If you already manage submissions as an author, you can also open your consoles via the same venue: Your consoles (OpenReview).

Important dates

Submission deadline: June 10, 2026 AoE

Decision notification: June 17, 2026 AoE

Publication and Archiving Policy

This workshop is non-archival. Accepted submissions will be invited for poster presentation and may be listed on the workshop website, but they will not be formally published in proceedings. Authors retain the intellectual property of their submissions and may submit extended versions of their work to other venues.

Paper awards

The workshop will recognize outstanding contributions with a Best Paper Award.

Topics of interest

Topics include, but are not limited to:

  • Representation-aware state estimation
  • Explicit and implicit representations for mapping and localization
  • Hybrid representations combining model-based and data-driven approaches
  • Representation choices for closed-loop decision making
  • Task-aware perception and mapping for planning, navigation, and exploration
  • Scalability and long-term consistency of representations in real-world environments
  • Evaluation of representations based on task-level performance

Keynote Speakers

Ayoung Kim
Associate Professor
Seoul National University
Teresa Vidal-Calleja
Professor
University of Technology Sydney
Maani Ghaffari
Assistant Professor
University of Michigan
Yixi Cai
Postdoctoral Fellow
KTH
Lantao Liu
Associate Professor
Indiana University Bloomington
Guoquan Huang
Professor
University of Delaware & Meituan
Timothy D Barfoot
Professor
University of Toronto
John Leonard
Professor
MIT

Organizers

Yingyu Wang
Postdoctoral Research Associate
University of Technology Sydney
Lan Wu
Lecturer
University of Western Australia
Tiancheng Li
Postdoctoral Research Fellow
University of Technology Sydney
Chuchu Chen
Assistant Professor
The George Washington University
Marija Popović
Assistant Professor
Delft University of Technology
Cédric Le Gentil
Researcher
ETH Zurich
Jaime Valls Miro
Professor
AZTI

Schedule

Time Session Speaker
8:30-8:35 Welcome & Introduction
8:35-9:00 Invited Talk 1 - Continuous Safety: Spatial Fields for Control, Planning, and Learning Teresa Vidal-Calleja
University of Technology Sydney
9:00-9:25 Invited Talk 2 - TBA Lantao Liu
Indiana University Bloomington
9:25-9:50 Invited Talk 3 - Computational Symmetry and Learning for Robotics Maani Ghaffari
University of Michigan
9:50-10:05 Early Career Talk - TBA Yixi Cai
KTH Royal Institute of Technology
10:05-10:30 Invited Talk 4 - Local Maps Are All You Need Timothy D Barfoot
University of Toronto
10:30-11:00 Poster Session & Coffee Break
11:00-11:25 Invited Talk 5 - Visual-Inertial Perception: From Kinematics to Dynamics Guoquan (Paul) Huang
University of Delaware & Meituan
11:25-11:50 Invited Talk 6 - From Geometry to Neural Fields: Representation Choices for Robust and Scalable LIDAR Autonomy Ayoung Kim
Seoul National University
11:50-12:15 Invited Talk 7 - Towards "Undo" for the Physical World John Leonard
MIT
12:15-12:20 Closing Remarks

Contact

Email: yingyu.wang@uts.edu.au

For inquiries about the workshop, submissions, or general information, please feel free to contact us.