ActionRoPE controls planar-camera game world models with world coordinates
What happened
Zhiyang Deng proposes ActionRoPE, a parameter-free action interface that converts camera displacement caused by actions into world-coordinate rotary positions. Tokens at the same world position keep the same positional encoding over time, which supports spatially consistent control in world models of planar-camera games. The paper reports that, compared with four representative action-conditioning mechanisms, the method follows the commanded actions more faithfully and reduces revisit drift by 66% relative to the strongest baseline. It also introduces Coordinate-Conditioned Noise Scheduling (CNS), which stabilizes observed regions earlier and generates new regions by referencing them. The same interface transfers to video backbones with 1.3B to 33B parameters and to out-of-distribution scenarios. All of these results are self-reported by the authors. The paper also notes that this representation fits planar-camera game worlds such as top-down and isometric views poorly, so its scope of applicability is limited.
AI-written from the coverage · updated 2026-10-21
Coverage timeline
Follow the coverage to see the story from different sides.
- arXiv Game World Models & GenerationPickActionRoPE: Using world coordinates for spatially consistent control in planar-camera game world models
ActionRoPE turns action-driven camera displacement into world-coordinate rotary positions, so tokens at the same world position keep the same positional encoding over time. The paper also introduces Coordinate-Conditioned Noise Scheduling (CNS), which stabilizes observed regions earlier and generates new regions by referencing them. Compared with four representative action-conditioning mechanisms, the method follows commanded actions more faithfully and cuts revisit drift by 66% relative to the strongest baseline. It also transfers to video backbones with 1.3B to 33B parameters and to out-of-distribution scenarios.
- arXiv Game World Models & GenerationPickActionRoPE enables spatially consistent camera-action control for planar-camera game world models using world coordinates
ActionRoPE is a parameter-free action interface that converts action-induced camera displacement into world-coordinate rotary positions, so tokens at the same world location keep the same positional encoding over time. The paper also proposes Coordinate-Conditioned Noise Scheduling (CNS), which stabilizes observed regions earlier so that new regions are generated with reference to them. Compared with four representative action-conditioning mechanisms, the method follows commanded actions more faithfully and reduces revisit drift by 66% relative to the strongest baseline. The same interface transfers to video backbones from 1.3B to 33B parameters and to out-of-distribution scenarios.