Conformal Orbit-Valid Trust Horizons for Equivariant World Models

arXiv 2026 56.5 method, theory

TLDR

Certifies trust horizons for equivariant world models using conformal prediction, showing orbit-constant rollout errors and zero violations in audits.

Reasoning

The paper introduces a novel conformal calibration method for trust-horizon certification in equivariant world models, with strong theoretical results on orbit invariance. However, the empirical evaluation is limited to symmetric 2D and 3D substrates, and the practical applicability to complex real-world scenarios remains unclear.

Read-first score

Read-first score 56.5, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 26.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.96995536

Methodology quality 18%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,evaluation,metric,result

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=code

Topical relevance 29%

37.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 389.

Keyword Scores

world model

world dynamics prediction

model-based reinforcement learning world model

world simulator

generative world model

interactive world model

video world model

Deep Analysis

Innovations

Conformal orbit-valid trust horizons for equivariant world models
Split-conformal multiplicative factor calibration of raw horizon curves
Structural result: exact equivariance transports calibrated trust-horizon curve over group orbit
Certificate-level calibration-cost study revealing two complementary regimes

Methodology

The method forms a raw horizon curve from a one-step latent residual and a finite-time expansion estimate, then calibrates it using a split-conformal multiplicative factor. Evaluation is performed on reproducible audit sets and stable audits across symmetric 2D and 3D yaw environments, comparing equivariant, plain, and augmented models. Metrics include violation rates, orbit-transport residuals, and certified-to-measured horizon ratios.

Key Results

The conformal factor γ_α=1.0 on the reproducible audit set, with zero anti-conservative violations across 50 stable audits (exact-binomial 95% upper bound 5.8%). Orbit-transport residuals are small (median 1.1%, max 4.1% over 14 orbit audits), and the certificate is non-vacuous (median certified-to-measured horizon ratio 0.67). On a 3D yaw audit, the equivariant model achieves a one-sector safe and non-vacuous orbit-valid certificate, while non-equivariant baselines incur violation, slack, sharpness, or additional-sector costs.

Limitations

The certificate is a conservative, distributional audit rather than a global reachability guarantee
Certificate-guided subgoal spacing is not confirmed in the current 3D CEM-MPC behavior layer

Resources

Paper