Driving · 2026-10-22
LLM-Based Exogenous World Modeling for Dynamic Scheduling
Ni Zhang, Haofeng Yuan, Jianan Zhou, Shiji Song, Zhiguang Cao
AI-assisted summary based on the paper. Consult the linked source for methods and limitations. Benchmark numbers appear after editor verification.
TL;DR
AWMP (Afterstate World-Model Planning) tackles dynamic scheduling by using a pretrained LLM as an exogenous world-model component that forecasts future job-arrival pressure via a compact early–late belief. Afterstate decomposition keeps deterministic shop transitions exact, while online residual self-calibration and surprise-triggered belief refresh adapt forecasts. Short scenario rollouts and meta-evolved planning yield lower total weighted tardiness than the best of eight classical priority-dispatching rules.
Why it matters
The authors address dynamic scheduling where future arrivals are uncertain and poor decisions can compound under nonlinear (weighted tardiness) objectives. AWMP separates known deterministic dynamics from uncertain arrivals, allowing planning with LLM-based exogenous forecasts while avoiding learning a full transition model or training a task-specific neural dispatch policy. Reported improvements are relative to strong classical baselines across multiple arrival distributions.
Method
- Afterstate decomposition: represent known deterministic shop effects exactly and isolate future job arrivals as the only exogenous uncertainty; the LLM forecasts only aggregated arrival pressure.
- LLM-based exogenous world model: predict a compact belief (numbers of arrivals in early and late horizon intervals) from observed arrival history; adapt via online residual self-calibration (prompt-level bias correction with frozen LLM).
- AWMP planning: belief-conditioned action scoring with top-K screening, fixed scenario generation from the belief, and short rollout evaluation; refresh LLM queries only when surprise-triggered, otherwise reuse cached belief across steps.
Limitation
AWMP represents future arrivals through a compact early–late workload belief rather than detailed job-level predictions (e.g., job attributes). The authors also note experiments are constructed from standard JSSP/FJSP benchmarks under four controlled arrival patterns, and that belief updates depend on LLM latency (more efficient forecasting models could help).
Abstract (from arXiv)
Dynamic scheduling requires dispatching online arrivals while anticipating future jobs, as poor decisions can compound downstream costs under nonlinear tardiness objectives. We show that a pretrained large language model (LLM) can serve as the exogenous component of a structured world model, forecasting uncertain future job arrivals. Specifically, we exploit an afterstate decomposition, which separates known deterministic shop transitions from uncertain exogenous job arrivals, enabling the LLM to forecast only the latter through a compact belief representation. Building on this, we develop Afterstate World-Model Planning (AWMP), which combines an evolved, belief-conditioned dispatching strategy with short scenario-based rollouts. A surprise-triggered refresh mechanism queries the LLM only when newly observed arrivals deviate sufficiently from previous forecast, and online residual self-calibration further conditions the LLM on its recent forecast errors to handle nonstationary arrivals. Across diverse arrival distributions, AWMP reduces weighted tardiness by 6-21% on dynamic job-shop and 22-46% on flexible job-shop benchmarks, relative to the best of eight classical priority-dispatching rules.
Related papers
- BeliefGraph-JEPA: Structured Latent World Models for Action-Conditioned Time Series
- ActTime: Action-Conditioned World Modeling for Time Series
- ANTE: Anticipate, Then Execute -- Learning Internal World Models for Coding Agents from Verified Beliefs
- AnyGoal: On World Model Generalization Beyond Expert Rollouts
- Relational Learning as World Modeling
Publication record
Bibliographic evidence describes where and when a record appeared. It does not validate scientific claims.
First public release: 2026-10-21 · Indexed by Pub-AI: 2026-10-22 UTC
LLM-Based Exogenous World Modeling for Dynamic Scheduling
Preprint posted: 2026-10-21 · Original record ↗
- arxiv
- 2610.29308
Explicit source versions (1)
- v1 · 2026-10-21
- Preprint available
Source date: 2026-10-21 · First detected: 2026-10-25 UTC · Historical baseline
Source evidence (1)
- Official evidence ↗ · atom:published (official metadata)
Abstract · arxiv
Dynamic scheduling requires dispatching online arrivals while anticipating future jobs, as poor decisions can compound downstream costs under nonlinear tardiness objectives. We show that a pretrained large language model (LLM) can serve as the exogenous component of a structured world model, forecasting uncertain future job arrivals. Specifically, we exploit an afterstate decomposition, which separates known deterministic shop transitions from uncertain exogenous job arrivals, enabling the LLM to forecast only the latter through a compact belief representation. Building on this, we develop Afterstate World-Model Planning (AWMP), which combines an evolved, belief-conditioned dispatching strategy with short scenario-based rollouts. A surprise-triggered refresh mechanism queries the LLM only when newly observed arrivals deviate sufficiently from previous forecast, and online residual self-calibration further conditions the LLM on its recent forecast errors to handle nonstationary arrivals. Across diverse arrival distributions, AWMP reduces weighted tardiness by 6-21% on dynamic job-shop and 22-46% on flexible job-shop benchmarks, relative to the best of eight classical priority-dispatching rules.
