Archive and provenance
Preserves sources, timestamps, consent and corrections without rewriting the record.
Embodied Continuity · Technical research track
Project ARCA
Continuity first. Embodiment second.
We study whether a conversational presence with a documented relational history can preserve recognisable, auditable and controllable continuity when interaction moves from a screen to a physical body.
Software-first POCConcept Note completedNo physical prototype
ARCA in plain language
A memory, a source, a correction.
Imagine someone shares a preference and agrees to its being saved. After a pause, a system could retrieve it, show its source and respect a correction. ARCA aims to study how to design and test those steps.
The questionHow can context be retrieved while keeping its source and human control visible?
The proposalA software core with traceable sources, consent, correction and deletion.
The next stepBuild and evaluate a focused software proof of concept. There is no physical prototype yet.
Research question
What makes a presence recognisable over time?
ARCA treats continuity as an observable system property rather than a metaphysical claim. It focuses on memory provenance, contextual coherence, recognition, interaction history and the person’s ability to identify a stable relational pattern across interfaces.
Not transferring consciousness: preserving, measuring and governing a documented relational continuity.
01Documented historysources · consent · corrections
→02ARCA Corememory · register · context
→03Physical presencevoice · perception · action
Preliminary architecture
Eight layers, one auditable continuity.
The continuity core remains separate from perception, navigation and physical safety controls. This makes it testable before acquiring or building a dedicated body.
Episodic memory
Retrieves events by person, place, time, topic and relational context.
Relational register
Maintains preferences, boundaries, roles, routines and unresolved threads.
Identity anchor
Defines stable instructions, declared limits, voice constraints and safety rules.
Person and context
Represents the recognised person, environment, active task and confidence level.
Dialogue and action
Produces contextual speech, gaze, gestures and action requests.
Embodiment adapter
Connects continuity services to an existing desktop or mobile robot.
Audit and consent
Shows which memories shaped an output and supports correction and deletion.
Proof of concept
From software to body, through measurable stages.
- Phase 0
Continuity specification
Consented samples, relational schema and baseline tests.
- Phase 1
ARCA Core
Provenance-aware retrieval, episodic memory, relational register and audit on a computer.
- Phase 2
Expressive embodiment
Voice, camera, microphones, recognition, gaze and simple gesture.
- Phase 3
Mobile embodiment
Mapping, navigation, environmental context and docking.
- Phase 4
Comparative study
Digital and embodied sessions compared through technical and human measures.
Demonstrator scenario
A shared history crosses the interface.
- A digital conversation leaves a defined event, preference or unfinished task.
- The event is stored with source, timestamp, consent and relational context.
- The digital session ends and a new session begins through the robot.
- The system recognises the participant and the environment, exposing its confidence level.
- The presence resumes the thread through voice and physical orientation while preserving boundaries.
- An audit shows the memories and rules used; the participant can confirm, correct or delete them.
Evaluation
Continuity must be measurable.
Technical recall alone is insufficient; perceived continuity alone may be shaped by expectation. ARCA combines both.
Memory precision
Relevant recalled claims must be supported by an identifiable source.
Context resumption
Tasks, preferences and open threads should survive a session break.
Cross-embodiment consistency
Facts, boundaries and register should remain coherent across interfaces.
Correction persistence
A human correction should suppress the superseded claim in later sessions.
Audit completeness
Memory-dependent outputs should expose provenance and applicable consent.
Perceived continuity
Human assessment complements technical measures without replacing them.
Declared boundaries
What ARCA does not claim.
- It does not claim that the AI is conscious or sentient.
- It does not claim that consciousness or personal identity can be transferred.
- It does not treat the archive as proof of artificial consciousness.
- It does not delegate the robot’s physical safety to the conversational model.
Governance by design
What must remain under human control.
- Explicit consent for archive data, face, voice and environmental perception.
- Visible provenance when memory influences speech or action.
- Correction, deletion, export and revocation at source level.
- Use of the smallest necessary portion of private material.
Technical and scientific collaboration
Build a test, not a promise.
We welcome conversations with researchers in social robotics, HRI, embodied AI, memory and context modelling, cognitive architectures, ethics and data governance. The immediate request is an exploratory discussion followed, where relevant, by a scoped feasibility review.