Contactless Intimate Body Gesture Translation Platform

Abstract

This document proposes the Contactless Intimate Body Gesture Translation Platform, an integrated architecture for translating contactless bodily motion captured through cameras, depth sensors, XR hand tracking, and similar input systems into embodied-intent events carrying relational meaning for an AI companion. Individual gesture classifiers are separated as interchangeable Gesture Attachments and connected to Gesture Candidate Events, arbitration and normalization, virtual body-zone mapping, consent and boundary decisions, companion-state transitions, predictive micro-reactions, multimodal response generation, and optional haptic output. The first case study, AIR MOMI / Air Knead, converts repeated grasping, releasing, and kneading motions performed in open air into candidate embodied-intent events. Without making physical haptic feedback a prerequisite, it evaluates Relational Reach: whether AI-companion-side facial, postural, vocal, and conversational responses allow the user to perceive that their embodied intent has reached the other party. Its primary evaluation target is felt_connection. The second case study, AIR AI PET, applies Air Hover, Air Stroke, Air Hold, Air Press, Air Release, and related Gesture Attachments to interaction with an AI pet represented on a display or in XR space. Embodied actions such as approaching, stroking, holding, light pressing, and release are synchronized with visual, motion, and vocal reactions on the AI-pet side and with optional mid-air haptic presentation. This case study evaluates Haptic Presence through presence, perceived_softness, multimodal_congruence, and related measures. The objective is not to fully reproduce a real body or physical touch. Instead, the Platform defines a shared architecture in which bodily input, AI-side state transitions, multimodal reactions, and optional haptic output are semantically and temporally coupled to support embodied interaction with digital entities. The specification further defines consent-aware Refusal Semantics, a Predictive Micro-Reaction Layer permitting only neutral acknowledgement before consent is resolved, privacy-aware abstract-event processing, MVP stages, evaluation metrics, and failure conditions. AI Usage Disclosure: Generative AI was used as a production aid in preparing this specification. Its primary uses included developing the initial ideas, organizing the specification structure, drafting text, examining counterarguments and boundary conditions, refining expression, and restructuring passages. The subject selection, technical concepts, adoption/rejection decisions, final structure, source verification, content validation, additions and revisions, and publication decision were made by the author, kurato. Generative-AI output was not treated as final text without review; material was incorporated only after selection, editing, and verification by the author. Accordingly, generative AI served to support the author’s production and review process, while final responsibility for the content and publication of this specification rests with the author.