Vai al contenuto

the usual neXt Immersive Media Research

From LLM Prompts to Runtime-Validated Immersive Experiences: Constraint-Guided Co-Authoring, Behavioral Validation, and Multisensory Orchestration

02 / 2026 System Paper / Preprint

Abstract

LLM-assisted authoring can accelerate immersive production, but a convincing output is not necessarily executable, behaviorally correct, or ready for public use. We present TholusOS, a system for co-authoring multi-user immersive experiences across dome projection, audio, lighting, participant phones, physical effects, media services, and a live scheduler. TholusOS separates creative generation from operational authority. A macro compiler derives an experience profile, temporal windows, a system-owned cue skeleton, and readiness requirements; the LLM fills only exposed creative fields. A micro compiler maps accepted moments through semantic anchors and modality lanes into synchronized runtime cues. Family-aware audits inspect progression, state transitions, action-consequence chains, service flow, and audience-perceptible outcomes. Cue timing contracts identify elements that are locked, narrowly adjustable, window-flexible, or optional, so repair remains proposal-only until review, application, and revalidation. Audience-readiness gates distinguish authoring, simulation, controlled participation, and production. We describe the implemented architecture, its uneven integration maturity, and a reproducible software snapshot, and define an evaluation protocol for authorial control, behavioral validity, multisensory timing, repair safety, and gate calibration. Executable Experience Grammar denotes a typed intermediate representation and transformation rules, not a formal language with proved soundness; liveReady is a software state rather than physical certification. This system report does not claim comparative or hardware-in-the-loop results not yet measured.

Keywords

  • LLM-assisted authoring
  • mixed-initiative authoring
  • behavioral validation
  • XR authoring
  • multisensory orchestration
  • runtime assurance
  • human-in-the-loop repair

How to cite

Riccio, D., & Guarnieri, E. (2026). From LLM Prompts to Runtime-Validated Immersive Experiences: Constraint-Guided Co-Authoring, Behavioral Validation, and Multisensory Orchestration (Version 5) [Technical report / preprint]. the usual neXt S.r.l. https://www.academia.edu/170244904/From_LLM_Prompts_to_Runtime_Validated_Immersive_Experiences_Constraint_Guided_Co_Authoring_Behavioral_Validation_and_Multisensory_Orchestration

BibTeX

@techreport{riccio_guarnieri_2026_tholusos,
  author = {Riccio, Dario and Guarnieri, Edoardo},
  title = {From LLM Prompts to Runtime-Validated Immersive Experiences: Constraint-Guided Co-Authoring, Behavioral Validation, and Multisensory Orchestration},
  institution = {the usual neXt S.r.l.},
  type = {Technical Report / Preprint},
  number = {Version 5},
  month = jul,
  year = {2026},
  url = {https://www.academia.edu/170244904/From_LLM_Prompts_to_Runtime_Validated_Immersive_Experiences_Constraint_Guided_Co_Authoring_Behavioral_Validation_and_Multisensory_Orchestration}
}