The robot is an environment, not a special chat shortcut#

Work Coordinator: admission and lifecycle of finite work#

The Work Coordinator is the system queue owner. User environment requests, Robot Operator children, and other finite producers submit work to it. Those producers do not start private coordinators.

Admission records why the work exists, who requested it, its priority, and the workflow handler that will execute it. Queue completion remains distinct from robot terminal feedback and semantic objective completion inside the workflow.

Environment Mode: semantic decision owner#

Environment Mode receives the current observation and bounded task envelope. Its context includes the connected session, advertised actions and robot commands, current state, fresh visual or audio material when available, prior task state, and request provenance.

One semantic selector produces a narrow result:

• a conversational response;
• a capability-bound action list;
• an optional bounded movement request;
• a task decision describing objective state, completion basis, and continuation.

The Action Parser validates and normalizes that output. It does not accept an action simply because the model emitted a familiar word.

Environment Task State: one bounded objective owner#

Task State prepares and reduces each Environment objective. It owns:

• objective identity and correlation;
• step and continuation bounds;
• pending action state;
• expected completion basis;
• correlated action and observation evidence;
• terminal outcome and the reason for it.

This prevents other nodes from creating their own competing validators or completion stores. The selector proposes; Task State decides whether the objective lifecycle can advance or close.

For visual goals, current behavior requires fresh frame-cited evidence. If the frame does not prove the stopping condition, a bounded continuation preserves the admitted action rather than substituting an unrelated motion or fabricating success.

Environment Bridge: external transport owner#

The singleton Environment Bridge agent owns the connection to the Ainekio adapter. It transports observations and actions, publishes connection state, and handles the bounded camera, microphone, and robot-speech envelopes.

It does not reinterpret Ainekio protocol details inside Environment Mode. The Ainekio adapter translates the generic action to the selected robot, waits for terminal robot feedback, and returns the normalized result.

A complete action turn#

1. Work Coordinator admits the Environment work.
2. Task State prepares the bounded objective.
3. Environment context binds the selector to current Ainekio capabilities.
4. The selector proposes a response, action, and task decision.
5. Action Parser validates the proposal.
6. Environment Bridge Out sends the correlated action.
7. Ainekio adapter translates and admits it to protocol v1.
8. The robot acknowledges, executes, and returns a terminal result.
9. Camera or state observations return with their own correlation.
10. Task State reduces the evidence and continues or closes the objective.
11. Conversation Buffer and TTS publish the user-facing response.

No single “robot done” message owns all eleven steps.

Conversation from the robot#

A VAD-bounded microphone utterance enters Environment Bridge as a typed audio envelope. MetaHuman Whisper produces the transcript. The Environment path admits it into the canonical Conversation Buffer rather than a robot-only chat history.

When a camera frame shares the utterance origin, the audio and visual material can be joined before the Environment turn. The response can then use the same Conversation Buffer and TTS owners as other MetaHuman interactions.

Speech back to Ainekio#

Environment speech uses the configured MetaHuman synthesis owner, creates a bounded PCM artifact, and transports it through Environment Bridge. Ainekio paces the artifact to the robot speaker and returns correlated playback status.

This preserves a clean division:

• MetaHuman owns provider choice, voice configuration, synthesis, and the speech artifact;
• Environment Bridge owns robot speech transport;
• Ainekio owns physical pacing, readiness, cancellation, I2S delivery, and terminal playback feedback.

What the current integration proves#

Source-level validation covers capability binding, action parsing, task lifecycle, visual completion bases, bridge envelopes, and robot adapter paths. Physical traces have shown reactive actions and at least one autonomous movement reaching terminal robot completion.

The integration is therefore implemented, not hypothetical. The remaining work is acceptance under real conditions: current rebuilt runtime, repeated transport stability, fresh-frame semantic proof, reliable voice, and multiple bounded autonomy cycles without stale or misleading output.