Physiological World Models for Human State Transitions

Chongyang Zhang; Rendong Wang; Hao Zheng; Hanwen Zhang; Yang Liu; Xiaolong Wei; Bin Chong

cs.AIAug 15, 2026

Physiological World Models for Human State Transitions

Authors: Chongyang Zhang, Rendong Wang, Hao Zheng, Hanwen Zhang, Yang Liu, Xiaolong Wei, Bin Chong

Organizations: Fullive-AI. · Peking University.

Abstract

Continuous multimodal sensing now allows human physiology to be observed throughout daily life rather than only during occasional clinical visits. However, most health artificial intelligence systems are designed to recognize current states, estimate risks or analyse individual biomarkers. They do not directly model how physiological states change in response to real-world events, behaviours, contexts and interventions. Here we propose the Physiological World Model (PWM), an event-conditioned framework for learning these changes at the level of the whole person. We introduce the HumanState Transition Token, a structured, quality-scored unit that connects the physiological state before an event with the event or action, relevant context and intervention information, the physiological trajectory after the event, observed outcomes and data quality. We describe four capability levels, from state representation to bounded intervention planning, together with four data acquisition and validation protocols. We also propose six benchmark tasks covering HumanState representation, forecasting across multiple timescales, individualized response prediction, simulation of alternative interventions, bounded planning and reliability under distribution shift. Together, this framework provides a practical path towards personalized health management, behavioural intervention design and clinician-supervised decision support, while clearly separating prediction from causal inference and making uncertainty, safety, governance and limits of use explicit.

Figures & tables

Figure 1: Physiological world models within a multiscale hierarchy of biological and health modelling. The proposed taxonomy orders modelling scopes from population and One Health systems [ 37 ] , through integrated physiology at the individual level, to tissue and organ [ 38 ] , cellular [ 39 ] , and molecular or biomolecular processes [ 40 ] . PWMs occupy the individual-level layer, whereas models at broader scales address population, cross-species, and environmental dynamics, and models at finer scales resolve processes within organs, cells, or molecules. The lower panel shows illustrative domain-specific specializations within the PWM level, including EEG, ECG/PPG [ 41 ] , sleep, metabolic, physiological-indicator, and disease-trajectory WMs [ 42 , 35 , 36 ] . These subdomains include both existing model instances and prospective directions. Arrows indicate the ordering of modelling scopes rather than causal or computational flow.

Figure 2: An illustrative functional architecture for a Physiological World Model. Event, action, HumanState, context, and intervention are encoded through specific encoders and taken as conditional input to a latent-state transition model; rollout branches represent alternative future trajectories and uncertainty, and task-specific heads map predicted states to observable physiological outcomes.

Figure 3: Capability ladder for Physiological World Models. Model capability increases vertically from L1 state representation through L2 state-transition prediction and L3 simulation and counterfactual rollout to L4 planning and intervention. The scope of temporal reasoning expands horizontally from the current state to single transitions, multi-step trajectories and closed-loop updating. The nested bands indicate the progressively broader functional scope of each capability level.

Figure 4: From fragmented digital-health records to protocol-based temporal alignment. A, Wearable, metabolic, clinical, behavioural, and environmental data are recorded asynchronously across devices and applications. B, First-person vision, audio, physiological signals, glucose, ECG, sleep, environmental context, activity, and clinical records are organized on a shared event-anchored timeline. Temporal alignment links events and interventions to pre-event baselines and subsequent physiological responses, enabling the construction of comparable HumanState Transition Tokens.

Task	Evaluation objective	Representative metrics	Capability and evidence requirements
T1: HumanState Representation and Estimation	Determine whether the model constructs a physiologically meaningful and transferable HumanState representation	Agreement with validated physiological measures; cross-device and cross-population transfer; robustness to noise and missing modalities	L1; primarily P1
T2: Multi-horizon State-Transition Forecasting	Predict how HumanState evolves across multiple timescales following a specified event, action, or intervention, conditional on the current HumanState and Context	Trajectory error; effective forecasting horizon; prediction-interval coverage; calibration	L2; primarily P2
T3: Event-Conditioned and Individualized Response Prediction	Determine whether the model captures person-specific response patterns and heterogeneity following a given event, action, or intervention	Response direction; peak amplitude; time to peak; recovery rate; personalization gain	L2; P2–P3
T4: Alternative Intervention Simulation	Compare plausible trajectories under alternative interventions	Held-out intervention error; intervention-ranking accuracy; cross-condition generalization error; uncertainty calibration	L3; primarily P3
T5: Bounded Planning and Steerability	Rank intervention options under explicit objectives, costs, contraindications, and safety constraints	Prospective outcome improvement; intervention efficiency; ranking accuracy; safety violations	L4; P3–P4
T6: Reliability under Distribution Shift	Determine whether the model recognizes when its predictions become unreliable	Robustness gap; calibration error; uncertainty–error association; abstention performance; subgroup failure analysis	Cross-cutting across L1–L4; P1–P4, with external or prospective evaluation under relevant distribution shifts

Table 2: Benchmark tasks for Physiological World Models.

Application horizon	Representative applications	Required tasks	Capability level	Evidence and validation requirements
Near-term	Sleep and fatigue monitoring; exercise recovery forecasting; metabolic stability monitoring; stress-related recovery	T1, T2, T3 and T6	Primarily L1–L2	P1–P2 data; external validation; personalization tests; calibration and applicability boundaries
Mid-term	Meal- and caffeine-timing guidance; sleep scheduling; exercise planning; work–rest adjustment	T1–T4 and T6	Primarily L2–L3	P2–P3 data; repeated or N-of-1 observations; crossover interventions; intervention-ranking validation
Long-term	Chronic disease trajectories; adaptive rehabilitation planning; clinical-trial enrichment; bounded clinical decision support	T1–T6	Primarily L3–L4	P3–P4 evidence; prospective validation; safety gates; clinical oversight; regulatory and accountability requirements

Table 3: Application horizons and indicative benchmark profiles.

World Models

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

See all 366 papers

Toward World Modeling of Physiological Signals with Chaos-Theoretic Balancing and Latent Dynamics
May 14, 2026Yunfei Luo, Xi Chen, Yuliang Chen +8Multivariate Time Series Forecasting World Model Learning
University of California San Diego · Dartmouth College · Amazon Web Services +3
PatientWM: Towards Generalizable Clinical World Models via Unified Patient States
Oct 12, 2026Ziwen Kan, Jie Peng, Bo Yan +5Action-Conditioned World Models Clinical Prediction
University of Science and Technology of China · University of Central Florida · Princeton University +2
Simulating clinical interventions with a generative multimodal model of human physiology
Apr 30, 2026Guy Lutsker, Gal Sapir, Jordi Merino +7Multivariate Time Series Forecasting Healthcare
Department of Computer Science and Applied Mathematics, Weizmann Institute of Science, Rehovot, Israel. · Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot, Israel. · NVIDIA, Tel Aviv, Israel. +5

May 14, 2026J/K move · Enter open · S save

Toward World Modeling of Physiological Signals with Chaos-Theoretic Balancing and Latent Dynamics

Yunfei Luo, Xi Chen, Yuliang Chen, Lanshuang Zhang+7

University of California San Diego · Dartmouth College · Amazon Web Services+3

Multivariate Time Series Forecasting World Model Learning Latent Dynamics Modeling Time Series Forecasting

Open paper AskPDF arXiv

Physiological time series signals reflect complex, multi-scale dynamical processes of the human body. Existing modeling studies focus on static tasks such as classification, event forecasting, or short-horizon next step prediction, while long-horizon signal-level forecasting and predictive nature of physiological signals remain underexplored. We introduce NormWear-2, a world model that encodes both multivariate physiological signals and clinical intervention variables into a shared latent space and models their joint temporal evolution as a dynamical system. Our approach combines inference from prior pre-trained knowledge (intuition) with instant non-parametric latent state transition adaptation (insight), enabling coherent forecasting across multiple temporal scales, conditioned on heterogeneous clinical interventions. During the pretraining phase, we find that chaos-theoretic balancing of dynamical regime diversity yields more robust representations, with a smaller balanced corpus outperforming one twice its size and capturing bifurcation regimes. We evaluate the world model performance across diverse real-world physiological datasets spanning heterogeneous temporal resolutions and intervention regimes, covering daily life, point-of-care, and clinical settings, including fitness planning, hemodialysis, diabetes management, and surgical monitoring. These evaluation datasets comprise records from 8,026 subjects, spanning study durations from 3.2 hours for high-resolution signal data to 2.3 years for longitudinal clinical biomarker tracking. NormWear-2 achieves the best overall forecasting performance across time, frequency, and latent representation domains, with significant improvements over state-of-the-art time series foundation models, while maintaining competitive downstream representation quality, providing a step toward general-purpose world models for physiological signals.

Abstract

World Models

Explore similar work

Toward World Modeling of Physiological Signals with Chaos-Theoretic Balancing and Latent Dynamics

PatientWM: Towards Generalizable Clinical World Models via Unified Patient States

Simulating clinical interventions with a generative multimodal model of human physiology