Behavior-Aware Anthropometric Scene Generation for Human-Usable 3D Layouts

Paper Title

Behavior-Aware Anthropometric Scene Generation for Human-Usable 3D Layouts

Publication Info

• Topic area: 3D scene generation with a focus on human usability and anthropometric constraints.
• Keywords: 3D scene generation, anthropometric design, vision-language models, human usability, spatial optimization, human-object interaction, XR environments, functional clearance, layout evaluation, human-centered design.

Background and Problem

• Problem / challenge: Existing 3D scene generation methods prioritize visual and semantic plausibility but fail to account for human usability, such as operational clearances and interaction zones tailored to individual anthropometric data.
• Significance: Usable layouts are critical for real-world applications, including XR environments, ergonomic assessments, and collaborative workspaces, where human movement and interaction fidelity are essential.
• Motivation and related work: Prior methods, such as LayoutVLM, rely on generic spatial constraints or LLM-derived common sense, which overlook individual body dimensions and behavioral patterns. This paper addresses the gap by integrating behavioral reasoning and anthropometric data into scene generation.

Solution

• Proposed approach: Behavior-Aware Anthropometric Scene Generation framework.
• Novelty:
  1. Integration of vision-language models (VLMs) to infer object-behavior relationships and spatial constraints.
  2. Grounding spatial constraints in individualized anthropometric data to ensure usability.
  3. Differentiable optimization of layouts using behavior-aware and anthropometric constraints.
  4. Comprehensive evaluation combining technical metrics, user perception studies, and real-scale usability tests.
• Procedure and key techniques:
  1. Semantic and Behavioral Representation: Analyze 3D assets to construct behavior-aware relational representations, linking object semantics, human-object interaction patterns, and group-level spatial relations.
  2. Constraint-based Layout Generation: Infer anthropometric constraints (e.g., clearance, reachability) and encode them as differentiable penalty terms for gradient-based optimization.
  3. Evaluation: Compare generated layouts with baselines using geometry-based metrics, expert perception ratings, and human usability studies in physical environments.

Results

• Concrete findings:
  • Improved task completion time: HO layouts reduced task times by 13.1%–21.1% compared to the baseline.
  • Enhanced trajectory efficiency: HO layouts reduced detours and unnecessary actions by 29.4%–46.7%.
  • Higher interaction-space utilization: HO layouts achieved up to 52.9% better volumetric occupancy ratios.
  • Geometry-based metrics: HO layouts improved in-boundary scores (+10.2%) but had slightly lower collision-free scores (−1.8%) compared to the baseline.
• Advantage over baselines:
  • HO layouts consistently outperformed both baseline and PO layouts in usability metrics, including functional usability, trajectory continuity, and operational clearance.
  • Expert perception ratings showed significant improvements in position appropriateness, orientation appropriateness, semantic plausibility, physical plausibility, and functional usability (median scores: HO = 5.0 vs. baseline = 3.0).
• Experiments / evaluation:
  • Geometry-based evaluation: Collision-free and in-boundary scores for 20 scenes.
  • User perception study: 16 experts rated 20 layouts across five criteria.
  • Individual usability study: 20 participants performed 10 tasks in office and lounge layouts under three conditions (Baseline, PO, HO).
  • Group usability study: 18 participants in six teams evaluated collaborative usability in shared layouts.
• Limitations and future work:
  • Limited validation scope: 20 scenes and 6 user study layouts.
  • Focus on horizontal arrangements; vertical interactions (e.g., shelf heights) were not optimized.
  • Multi-user optimization prioritized maximum body dimensions, occasionally creating discomfort for smaller users.
  • Generalization to unconventional furniture designs and XR environments remains unexplored.

Summary

This paper introduces a Behavior-Aware Anthropometric Scene Generation framework that integrates vision-language models and anthropometric data to optimize 3D layouts for human usability. By addressing operational clearance and interaction zones, the framework ensures layouts accommodate dynamic human actions beyond static collision avoidance. Results from technical validation, expert perception studies, and user usability tests demonstrate significant improvements in task efficiency, trajectory continuity, and interaction-space utilization compared to baseline methods. The approach has potential applications in XR environments, embodied AI, and digital twin systems, though further research is needed to generalize findings to diverse contexts and vertical interactions.