Latent Dynamics · 2026-10-20
Operator-Mismatch Contrastive Learning for World-Relative Trajectory Coherence
One Octadion, Firmansyah Sundana, Novanto Yudistira
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
The authors formalize “operator-blindness” in world-relative trajectory scoring: a scorer can reject corrupted futures and still order current-world vs other-world valid futures at chance, and standard generic top-1 evaluation can hide this. They propose Operator-Mismatch Contrastive (OMC) learning using valid-but-wrong-operator rollouts as semantic hard negatives. On synthetic IID/OOD and CoPhy feature-level validation, OMC improves operator-sensitive coherence scoring while keeping generic top-1 near 0.99.
Why it matters
Ranking plausible futures is insufficient if the scorer is insensitive to which latent dynamics/operator governs the present context. OMC directly trains/evaluates “world-relative coherence” by inserting wrong-operator but valid trajectories into the contrastive candidate set, and by using an operator-sensitive protocol that can detect operator-blindness. This provides an auxiliary trajectory-coherence signal that can be attached to existing trajectory-scoring representations rather than requiring end-to-end generative modeling or planning.
Method
- Define operator-blindness: operator-conditional coherence fails when current-operator valid futures are not ordered below valid wrong-operator futures, even after rejecting incoherent noise.
- Use OMC: construct semantic hard negatives τ_wrong as valid trajectories from a different operator, and train an energy scorer with an InfoNCE-style ranking loss over {τ_same, τ_state, τ_temporal, τ_wrong, τ_rand}, including KL and stability terms.
- Evaluate with operator-sensitive metrics (e.g., law_pair, same_lt_wrong, full_stratification, energy-matrix retrieval) paired with a matching protocol that inserts τ_wrong into the comparison.
Limitation
OMC is a minimal, attachable coherence signal rather than a standalone world model: it improves discrimination of world-relative coherence, not generation, planning, or control. It also requires access to alternative-operator trajectories at training/evaluation time to construct τ_wrong; wrong-operator rollouts are not required at inference.
Abstract (from arXiv)
World models are increasingly judged by how well they rank plausible futures, yet plausibility alone is the wrong target. The same scene can evolve under many latent rules, and a future that is smooth and physically valid under one rule can be incoherent under the rule actually in force. We formalize the resulting failure, operator-blindness, and show that it is systematically hidden by standard evaluation: a scorer can attain near-perfect generic ranking while separating current-world from other-world futures no better than chance. We introduce Operator-Mismatch Contrastive (OMC) learning, whose semantic hard negatives are valid-but-wrong-operator trajectories, and a matching protocol that measures world-relative coherence directly. On a controlled synthetic benchmark with in- and out-of-distribution operator parameters, across two trajectory-scoring backbones, and on feature-level external validation with CoPhy ballsCF, OMC raises operator-sensitive scoring far above a chance baseline that nonetheless achieves 0.99 generic top-1. We prove that the wrong-operator negative is necessary both for the protocol to detect the failure and for the objective to fix it, and position OMC as an auxiliary coherence signal (a trajectory-coherence scorer, not a generative model, planner, or control agent) that can be attached to existing world-model representations.
Related papers
- WorldFoundry: Towards Unified Infrastructure for World Intelligence
- Seeing Beyond the Next Step: World-Model-Guided Human-Like Navigation in Multi-Agent Scenes
- The Posterior-Prior Epistemic Gap: When World Model Imagination Fails to Carry Task-Relevant Signals
- HiFOR: Hierarchical Foresight with Action-Guided World Grounding for Vision-Language-Action Model
- FutureWorlds: Learning Robotic World Models from Alternative Futures
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-18 · Indexed by Pub-AI: 2026-10-20 UTC
Operator-Mismatch Contrastive Learning for World-Relative Trajectory Coherence
Preprint posted: 2026-10-18 · Original record ↗
- arxiv
- 2610.25360
Explicit source versions (1)
- v1 · 2026-10-18
- Preprint available
Source date: 2026-10-18 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
World models are increasingly judged by how well they rank plausible futures, yet plausibility alone is the wrong target. The same scene can evolve under many latent rules, and a future that is smooth and physically valid under one rule can be incoherent under the rule actually in force. We formalize the resulting failure, operator-blindness, and show that it is systematically hidden by standard evaluation: a scorer can attain near-perfect generic ranking while separating current-world from other-world futures no better than chance. We introduce Operator-Mismatch Contrastive (OMC) learning, whose semantic hard negatives are valid-but-wrong-operator trajectories, and a matching protocol that measures world-relative coherence directly. On a controlled synthetic benchmark with in- and out-of-distribution operator parameters, across two trajectory-scoring backbones, and on feature-level external validation with CoPhy ballsCF, OMC raises operator-sensitive scoring far above a chance baseline that nonetheless achieves 0.99 generic top-1. We prove that the wrong-operator negative is necessary both for the protocol to detect the failure and for the objective to fix it, and position OMC as an auxiliary coherence signal (a trajectory-coherence scorer, not a generative model, planner, or control agent) that can be attached to existing world-model representations.
