The 2nd Workshop on Embodied Spatial Reasoning

Embodied spatial reasoning is the ability to interpret the positions, orientations, physical properties, and temporal dynamics of an agent and its surrounding objects in 3D space: a foundation for spatial understanding, reasoning, and generation in modern embodied AI and world modeling.

Yet work on this topic remains fragmented. Embodied agents derive spatial reasoning from multimodal alignment, while world models extract priors from video, with no shared interface for spatial, physical, and temporal modeling. This workshop brings together researchers across embodied AI, spatial reasoning, and world modeling to tackle object permanence, spatial memory, physical consistency, and long-horizon temporal coherence.

By bridging these communities, we aim to advance agents and video world models that maintain object permanence and spatial memory while preserving physical and temporal consistency in 3D space.

The paper track welcomes long and short submissions on spatial understanding, reasoning, and generation for embodied agents and world models: new methods, benchmarks, analyses, and position papers. We uphold high standards of review, with impartial and constructive feedback for every submission.

The demo track is a dedicated venue for novel, interactive, and visionary systems. Where the paper track emphasizes rigorous evaluation, the demo track focuses on active embodied agents and physically-grounded world models: systems you can watch behave in the real world. It shares the same reviewing process and the same deadlines as the paper track.