Dual-Constrained Co-Planning of Virtual Health Avatars and User Interfaces for Older Adults with Chronic Diseases

Digital health systems for older adults with chronic diseases must support task completion, risk communication, and emotional engagement. However, virtual avatars and user interfaces are usually generated separately, which can weaken the consistency between avatar behavior, interface hierarchy, and task risk. This study proposes a dual-constrained co-planning method that integrates continuous valence–arousal–dominance affective targets with task engineering constraints, including functional priorities and age-friendly boundaries. User needs are translated into three representative health-management tasks and encoded in a multi-source constraint graph. A two-stream conditional Transformer then jointly plans avatar behaviors and interface layouts, followed by constraint checking and automatic rollback. Across the three evaluated health-management scenarios, the framework produced task-sensitive changes in avatar expression, gaze, gesture, and interface priority and supported rule-based checking and rollback of predefined age-friendly constraints. Exploratory external comparisons and ablation analyses provided preliminary evidence regarding the roles of affective constraints, graph-based task representation, and cross-stream coordination in prototype-level affective matching, usability, constraint compliance, and avatar-UI coordination. The framework is intended as a structured approach for verifiable early-stage prototype generation and candidate screening rather than for direct clinical deployment.