Accurate, informative, and scalable world representations are an essential component of highly autonomous mobile robots and have been an important topic of research for several decades. As robots become more capable, deploying in larger and more dynamic, varied environments, requirements for such representations have grown apace. Handling multiple data modalities, levels of abstraction, and types of information (metric, topological, semantic, objects, etc.) remains challenging — even more so in so-called lifelong settings where robots must maintain world models over extended periods of time. Over the last forty years, roboticists have used techniques from many machine learning and statistics paradigms for mapping. However, none have been nearly as transformative as deep learning, and we are now at an inflection point in the pace of adoption and proliferation of deep learning techniques for representing models of the world suited to robotics. Such a moment offers an opportunity for retrospection: to consider lessons from previous eras of research that have stood the test of time, to carry such lessons forward into an age of research dominated by models relying on latent representations, and to understand in hindsight the limits and blind spots of previous paradigms. We may look forward as well: to understand the tradeoffs presented by newer learning and representation techniques, to share and discuss new examples of state-of-the-art technical approaches for robotic mapping and modelling, and to develop a shared view of the new frontier of challenges facing such systems as they are deployed in ever more challenging domains.

With these goals and challenges in mind, we extend a call for the following types of papers for submission to the Workshop on Retrospective and Future World Representations at IROS 2024:

Technical papers (up to 6 pages). Technical papers propose new methods for mapping or new world representations that represent the state-of-the-art in some aspect. Specific topics could include novel:

• Time-aware predictive world models;
• Action-conditioned world representations;
• Methods for updating maps over many deployments or long time periods;
• Hierarchical representations of environments;
• Methods that explicitly support structure learning in the environment;
• Methods for representing different information (e.g. spatiosemantic maps)” to “Methods for representing different types of information (e.g. spatial, temporal, semantic);
• Methods for memory-efficient map storage;
• Mapping primitives (e.g. Gaussian splatting, voxels, planes, etc.);
• Mapping data structures (e.g. oct-trees, place graphs, etc.).

Retrospective papers (up to 8 pages). Retrospective papers are survey papers or other studies examining existing research at scale. They do not have to be exhaustive but should provide a solid foundation for a more comprehensive survey or tutorial article. Such papers may taxonomize previous mapping research, identify important lessons from previous eras of research that have stood the test of time, discuss how to carry such lessons forward into an age of research dominated by models relying on latent representations, or provide insight into the limits and blind spots of previous paradigms.

Mini retrospectives (up to 4 pages). Mini retrospectives are unique to this workshop and analyse how 1 or 2 related papers that are at least 5 years old have changed the way we think about world representations. Their focus is narrow, with an emphasis on detailed studies of a single theme.

Position papers (4-8 pages). Position papers look to the future and argue for the importance of certain techniques, questions, evaluation paradigms, design philosophies, etc. in future research. They may also elaborate on particular tradeoffs presented by newer learning and representation techniques or develop a view of new difficulties facing state-of-the-art systems as they are deployed in ever more challenging domains.

This workshop is non-archival, and thus we are happy to review submissions that are concurrently under review elsewhere or have already been published in whole or in part.