Human-Oriented World Model

Saihui Hou; Zengyun Mou; Chunshui Cao; Yongzhen Huang

cs.AINov 2, 2026

Human-Oriented World Model

Authors: Saihui Hou, Zengyun Mou, Chunshui Cao, Yongzhen Huang

Organizations: Beijing Normal University · WATRIX AI

Abstract

Supporting a person over repeated interactions requires interpreting their current state in context and anticipating its subsequent evolution. We formulate the Human-Oriented World Model (HOWM) as a conceptual framework centered on task-relevant human states and their evolution. HOWM connects current observations and individual history with environmental and social context to organize continuing state estimation, personalized dynamics prediction, and action support. Dynamic expectations and new observations jointly inform the current estimate, which in turn anchors personalized forecasts based on shared knowledge and individual experience. An external policy relates forecasts under alternative conditions to the person's goals, preferences, and constraints when considering support. Actual responses and outcomes provide evidence for revising the state estimate and personal dynamic patterns, thereby informing subsequent estimation, prediction, and support. We illustrate the framework through hypothetical training and learning scenarios. Our research agenda asks whether accumulated interaction experience improves subsequent estimation and prediction, and whether using the model helps the person pursue their goals over time.

Figures & tables

Panel of Figure 1 — Figure 1: Functional organization of HOWM. State and Dynamics connect continuing estimates to personalized futures using history and context. An external policy relates candidate futures to the person’s goals and selects support. Actual interaction returns evidence for state revision and learning personal dynamics, which can inform subsequent estimation, prediction, and support. Arrows indicate information dependencies and allow shared or iterative computation.

Symbol	Meaning and role
$\mathcal{I}_{t},\ q_{j,t}$	Identifiers and uncertain assignments of observations or references to individuals.
$Z_{t}=(H_{t},E_{t},R_{t})$	Task-relevant human, environmental, and relational states being modeled.
$b_{t},\ b_{t}^{i}$	System estimates of the joint and individual states, including uncertainty.
$\mathcal{O}_{\leq t},\ \Delta\mathcal{O}_{t}$	Evidence available by time $t$ and newly available evidence at the current update.
$C_{t}^{i}$	Available information or estimates about the person’s environmental and social context.
$M_{t}^{i}$	Retained evidence and summaries with sources, times, and applicability conditions.

Table 1: Core notation and information roles in HOWM.

Focus	Question and comparison	Assessment targets
Continuing estimation	Does the estimate follow change? Vary state propagation, context, or personal history (H1) under common observations.	Successive estimates, uncertainty, and recovery after gaps or task changes, using appropriate state references.
Personalized dynamics	Does the model anticipate individual development? Compare shared, personal-baseline, and temporal-pattern predictors; examine historical contributions (H2).	Future change and trajectories at common origins and horizons, with starting-state information and contextual coverage declared.
Action support	Does using forecasts help the person? Compare support policies and separately examine action-conditioned prediction (H3).	Forecast quality under realized conditions; choices, burden, and outcomes relative to the person’s goals.
Continued updating	Does experience improve later use? Compare adapted and retained personal patterns on common later observations; assess corrections.	Later estimation and prediction, adaptation to change, recurring errors, and usefulness of subsequent support.

Table 2: Capability questions and corresponding validation designs. H1–H3 investigate information contributions within this agenda; continued updating is assessed through later use across the three capabilities.

Study focus	Required evidence	Validation targets
Continuing estimation and initial prediction	Repeated observations, context, activities, times, state references or later outcomes.	Successive estimates and forecasts followed by new evidence and updating; simple baselines and H1 comparisons.
Personalized dynamics	Individual trajectories across activities or sessions, relevant conditions, and outcomes.	Prediction of change beyond personal baselines; H2 historical contributions and adaptation with experience.
Conditional prediction and assistance	Proposals, human choices, executed actions, outcomes, goals, and observation conditions.	H3 predictive contribution; separate policy comparisons and human outcome assessment.
Learning and revision across stages	Successive model and information versions, interaction feedback, corrections, later observations, and references.	Later estimation, prediction, and support; adaptation and recurring errors, with original forecasts preserved.

Table 3: Validation pathways and required data. Stages may overlap; updating is assessed in each capability.

Modeling role	Sports and training	Education and learning
State and evidence	Movement control and training condition inferred from movement and reports.	Understanding and difficulty inferred from task responses and learner feedback.
Individual history	Associated practice trajectories, cue schedules, reports, and personal baselines.	Linked task sequences, hints used, feedback, and performance on comparable material.
Context	Exercise demands, court conditions, coach or partner roles, and cue pace.	Problem demands, instructions, and teacher or peer assistance.
Personal dynamics	Timing and stability in adapting to cue schedules, with applicability and uncertainty.	Progress across supported and independent tasks, with applicability and uncertainty.
Conditional futures	Movement trajectories and completion under the proposed practice arrangement.	Performance over a task sequence with specified assistance conditions.
Goals and support	Consistent footwork at manageable effort informs pace changes or clarification.	Independent problem solving guides consideration of demonstrations, hints, or further attempts.

Table 4: State, personal dynamics, and continuing use in the hypothetical training and learning settings.

World Models

World models represent and predict how environments change, helping agents plan, act, and reason about possible futures.

See all 391 papers

Mental World Modeling
Jul 29, 2026Hao Fei, Yiran ZhaoWorld Model Learning Theory of Mind
University of Oxford · National University of Singapore
World Models for Embodied Intelligence: From Plausible to Controllable to Actionable
Sep 15, 2026Nanjie Yao, Hao Wang, Chong Cheng +10World Models Action-Conditioned World Models
HKUST(GZ) · University of Oxford · Princeton University +8
From World Models to World Action Models: Rethinking Next-State Prediction
Sep 28, 2026Tingyu Yuan, Ziming Ji, Biaoliang Guan +9Robot Dynamics Learning Action-Conditioned World Models
CASIA · UCAS · BUPT +4

Jul 29, 2026J/K move · Enter open · S save

Mental World Modeling

Hao Fei, Yiran Zhao

University of Oxford · National University of Singapore

World Model Learning Theory of Mind Human Behavior Prediction Human Behavior Modeling

Open paper AskPDF arXiv

World models enable a predictive substrate for planning and action, yet existing formulations merely answer a physical question: what/where it is, and how will it evolve. Human behavior, however, is driven by hidden mental state (what a person believes, wants, intends, feels, and considers socially permissible), so a model that tracks the physical scene but not what each agent knows and believes about it predicts the wrong action for the right-looking scene. We formulate Mental World Modeling (MWM), a generic theoretical framework that makes mental variables core components of a world model rather than posthoc rationales: MWM aintains a coupled physical-mental world state, renders a target-specific partial observation, and simulates how candidate actions jointly update both components. We instantiate the framework in MENTIS, a training-free and fully inspectable baseline that decomposes the process into state parsing, target-observation generation, action decomposition, coupled physical and mental transition, and branch-level value evaluation. On a manually constructed, quality-controlled dataset of situated decision scenarios spanning text, image, and sounding-video stories, experiments with 8 modern LLM-based world models demonstrate that explicitly modeling the mental state is essential for predicting human decisions. Deeper analyses further expose the bottlenecks of current mental world modeling. We expect MWM as a next stage of world modeling, from simulating physical scenes to simulating the minds that act in them.

Abstract

World Models

Explore similar work

Mental World Modeling

World Models for Embodied Intelligence: From Plausible to Controllable to Actionable

From World Models to World Action Models: Rethinking Next-State Prediction