Oxford Team Proposes Oneira: Using Coding Agents to Maintain World State Tables for Precise Object Interaction and Long-term Consistency in Video Generation

1 reports archived1 independent sourcesupdated 10/13/2026, 12:00:00

Synthesis & Latest Updates

1 Sources Cross-Validated

Oneira introduces an explicit 'World State Table' maintained by a Coding Agent to address target misalignment and state loss during camera movements in existing video world models. This architecture upgrades open-ended generation to open-world interaction, ensuring that object modifications persist across subsequent frames, significantly improving interaction accuracy and temporal consistency.

LATEST/Oneira introduces an explicit 'World State Table' maintained by a Coding Agent to address target misalignment and state loss during camera movements in existing video world models. This architecture upgrades open-ended generation to open-world interaction, ensuring that object modifications persist across subsequent frames, significantly improving interaction accuracy and temporal consistency.

HEAT TRENDHourly heat curve

2 fully observed hours

Fewer than 3 fully observed hours; trend not plotted yet.

TIMELINECoverage timeline

Total 1 reports · Latest first

机器之心 (微信公众号)T2·Oct 13, 12:0078 pts
Oxford Team Proposes Oneira: Using Coding Agents to Maintain World State Tables for Precise Object Interaction and Long-term Consistency in Video Generation
- Key breakthrough lies in using a Coding Agent to maintain an explicit World State Table recording objects and their states, rather than relying solely on implicit features.
- Solves the pain point where traditional video models fail to preserve object interaction states (e.g., position, shape changes) after the camera moves away and returns.
- Enables a leap from 'navigation/simple interaction' to 'precise object manipulation + long-term consistency', supporting accurate modification of existing or newly generated objects.
View Evidence Original Source