Second Thoughts: Exploring and Planning through Comparative Imagination in World Models
We introduce SDI (Sibling Disagreement in Imagination), which connects these decisions through a shared actor-based reference.
ProofPaper ↗
Key points
- World models let agents compare imagined outcomes before acting, but an agent must still decide which alternatives to explore and when a proposed plan should replace its learned policy.
- SDI compares action sequences imagined from the same model state, using repeated rollouts of the actor's own sequence to measure variation under fixed behavior.
- Component ablations and robustness analyses further examine the roles of exploration, planning exposure, and action acceptance.
- Together, these results support comparing imagined alternatives against a shared actor reference as a unified approach to exploring and planning with world models.
Sources (1)
- [1]Second Thoughts: Exploring and Planning through Comparative Imagination in World ModelsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 06:12 PM
We introduce SDI (Sibling Disagreement in Imagination), which connects these decisions through a shared actor-based reference.
World models let agents compare imagined outcomes before acting, but an agent must still decide which alternatives to explore and when a proposed plan should replace its learned policy.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 20, 2026OverSeer: Multi-Agent World Models Beyond What Agents See
- Oct 19, 2026EmbodiedWorld-200K: Toward Embodied World Agents via Embodied-Planning Dataset and Interactive World Models
- Oct 19, 2026Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience
- Oct 19, 2026Agent World Model in the Wild: Computer-Use Agent Environment Synthesis
- Oct 19, 2026AnyGoal: On World Model Generalization Beyond Expert Rollouts
- Oct 19, 2026PVBench: Quantifying Physical Plausibility in Video Generation with Object-Level Physical Measurements