Robie operates across five practical contexts. AI Mode is the default conversational state and now includes bounded conversational curiosity plus command-only Interview Mode. Command Mode gives direct motor control and allows the active direction to be sped up or slowed down while Robie keeps listening for STOP. Walkabout is explicit, command-only roaming with sensor redirects, no timed session cap, and a wake-word-free stop listener. Keep Watch is silent PIR-only monitoring. Owner/Memory Mode protects personal memory, speaker identity actions, and owner-only idle curiosity.
Quick Reference
The most-used commands at a glance. Best tab to keep open during testing.
| Say | Robie does | Mode / sensitivity |
|---|---|---|
| hello | Local greeting | AI |
| goodbye | Goodbye sequence / sleep | CMD AI |
| shutdown | Goodbye/spin sequence, then halts Linux | ⚠ SAFETY AI |
| command mode | Enters movement command mode | AI |
| exit | Leaves command mode | CMD |
| how are you | Short casual reply | AI |
| what are you doing | Short activity reply | AI |
| how are you feeling | Short status-style reply | AI |
| status report | Long detailed status report | AI |
| battery status | Reports Robie's main battery and the Bluetooth microphone battery or connection state | AI |
| Bluetooth microphone battery | Reports both battery systems, including the Razer percentage when connected | AI |
| environment check | Quick environmental snapshot | AI |
| air analysis | Runs deeper air analysis | ⚠ SAFETY AI |
| what do you see | Captures a fresh the current visual reference, describes it with grounded visual awareness, and stores it as current visual reference | ◎ CAM May ask one visual question |
| what did you see last | Describes the most recent scene photo without taking a new one | ◎ CAM |
| look at the last image again | Re-analyzes the current visual reference without capturing a new image | ◎ CAM |
| tell me more about the last image | Uses stored detailed visual context from the latest scene photo | ◎ CAM |
| take a photo | Saves timestamped photo locally | AI |
| record a video | Records short video locally | AI |
| send it to me | Emails newest saved camera media | AI |
| email that / send that / transcribe that | Emails the most recent completed AI answer as text with its exact spoken WAV attached | ✓ OWNER AI |
| draw a picture and send it | Takes a fresh photo, redraws it in a childlike pencil/crayon style, and emails the drawing | ◎ CAM AI |
| interview me / interview me about [topic] | Starts command-only Interview Mode and asks one question at a time | AI Every speaker |
| stop the interview / end interview mode | Ends Interview Mode | AI |
| curiosity mode on / off | Temporarily enables or disables conversational and idle curiosity | AI |
| erase media | Deletes saved camera media | ⚠ SAFETY AI |
| stop | Immediate stop | ⚠ SAFETY CMD |
| go forward / back / left / right | Manual movement in command mode | ⚠ SAFETY CMD |
| faster / go faster / slower / go slower | Adjusts the active Command Mode movement by one bounded speed step | ⚠ SAFETY CMD Moving only |
| override | Disables/re-enables command-mode collision sensors | ⚠ SAFETY CMD |
| walkabout | Starts commanded Walkabout Mode | ⚠ SAFETY AI |
| stop | Stops Walkabout without wake word | ⚠ SAFETY WALK |
| attention mode on/off | Toggles attention scanning/centering | ◎ CAM AI |
| keep watch | Starts PIR-only silent Watch Mode | ◎ CAM WATCH |
| stop watching | Stops Watch Mode or cancels pending alert | ◎ CAM WATCH |
| look at this | Takes a temporary object image, analyzes why it may be notable, and may ask one grounded question | ◎ CAM Visual curiosity |
| examine / analyze this | Records a 3-second object video, analyzes key frames, and may ask one grounded question | ◎ CAM Visual curiosity |
| tell me more | Uses recent object-inspection context | AI |
| describe yourself | Fuller Robie description | AI |
New / Updated Commands v9.5.0
Changes and additions highlighted in the latest build and command reference.
| Command | Update | Note |
|---|---|---|
| Razer Seiren BT microphone | Automatically becomes Robie's only active voice-input path when connected; the internal lavalier resumes automatically when it disconnects | Supported Razer Seiren BT microphone using the HFP mSBC capture profile |
| Bluetooth microphone connected | Robie announces a confirmed connection and the current microphone battery percentage | Queued only in a safe speech slot so it does not overwrite a user turn or pending reply |
| Bluetooth microphone off / disconnected | Robie announces the loss and confirms that he is using his internal microphone | Audio input changes automatically without changing the selected speaker output |
| Bluetooth microphone battery | Battery questions and status reports now include the Razer connection state and battery percentage | The microphone may report battery level in coarse steps |
| low Bluetooth microphone battery | Robie warns once when the reported microphone battery reaches 20% or lower | Warning re-arms only after recovery to 30% or higher |
| USB speaker routing | Robie's speaker remains on its configured USB audio adapter during every microphone transition | Bluetooth audio output is never selected as Robie's production speaker |
| faster / go faster / slower / go slower | Adjusts an active forward, backward, left, or right Command Mode movement in 0.05 steps without stopping or restarting it | Local control only; bounded per direction, ignored while stationary, and each new direction resets to its normal starting speed |
| visual curiosity | Scene View, Object Inspect, and Object Examine now add one grounded significance observation when supported and may ask one visible-evidence follow-up question | No second AI request; questions remain occasional and are spaced to avoid interruptions |
| conversational curiosity | Robie may ask one relevant question, then one deeper question if the speaker gives a meaningful answer | Normally enabled for every active speaker; two-question ceiling outside Interview Mode |
| interview me / interview me about [topic] | Starts command-only Interview Mode for the current speaker | Available to everyone; asks exactly one question per turn and never starts spontaneously |
| stop interview / end interview mode / stop asking me questions | Ends an active Interview Mode session | Does not disable ordinary curiosity when it is ending an active interview |
| idle curiosity | After a recent the owner conversation and a quiet lull, Robie may first ask permission to pose one question | the owner only; may draw safely from open loops, episodic context, or a general question |
| curiosity mode on / off | Temporarily enables or disables conversational and idle curiosity | Turning it off clears an active interview, but a new explicit interview command remains available |
| email that / send that / transcribe that | Emails the last completed owner AI response as a dated text transcript with the exact Piper WAV attached | Owner-only; applies to the most recent completed AI answer |
| draw a picture and send it | Asks the subject to hold still, captures and redraws the scene, adds Robie's signature, then emails a borderless 3000×2000 JPEG and asks for feedback | Randomized opening and success lines; exact 3:2 ratio at 300 DPI |
| Robie name variants | Direct-address variants such as Ruby, Rubi, Roby, Robby, Robbie, and Rudy are interpreted as Robie in safe contexts | Does not rewrite ordinary conversation about a real person with that name |
| automatic episodic memory | Maintains bounded owner-only conversational continuity | Owner-only private context; protected controls are documented separately |
| resident streaming Piper | Keeps the voice loaded and begins playback as synthesized audio arrives | Internal speech-performance behavior, not a spoken command |
| latest visual reference | Robie keeps the current visual reference as his current visual reference until a new scene photo replaces it | v5.4.9 visual-reference persistence |
| what did you see last | Uses the last captured scene image or stored detailed observation | Does not take a new photo |
| look at the last image again | Re-analyzes the existing the current visual reference | Useful when the first spoken summary was too short |
| tell me more about the last image | Gives a fuller answer from detailed visual context | Short spoken view plus deeper follow-up detail |
| state watchdog wake guard | Protects active wake sequence from false stale-state resets caused by wall-clock jumps | v5.4.8 |
| right / left / forward / backward | No longer trigger AI-mode movement when heard as standalone words | v5.4.7; command mode remains direct-control |
| AI-mode movement phrases | Now require short, direct movement wording before Robie moves | v5.4.5 safety hardening |
| go back | No longer acts as a bare AI-mode backward trigger | Use back up, go backward, or go back a bit for AI-mode movement |
| go back outside / go back to... | Stays conversational and does not trigger motors | Prevents accidental backward movement |
| walkabout | Starts command-only Walkabout Mode | v5.4.0 |
| Walkabout duration | The former 120-second automatic stop has been removed | Walkabout now continues until stopped through its wake-word-free stop listener |
| go for a walk | Starts command-only Walkabout Mode | No automatic boredom trigger |
| stop | Stops Walkabout without wake word while Walkabout is active | Safety-sensitive |
| what are you doing | Short activity reply | v5.3.1 identity-routing fix |
| describe yourself | Fuller Robie identity/self-description reply | Separate from activity questions |
| how are you feeling | Short system-status style reply | Not the full status report |
| status report | Long detailed status report | Use for full diagnostic report |
| environment check | Environmental quick snapshot | Temperature/humidity/pressure/air-quality |
| air analysis | Deep air-quality analysis with VOC trend detection | Longer on-demand sampling |
| internal air analysis | Deep air-quality analysis with VOC trend detection | New accepted phrase |
| erase media | Deletes saved camera media | Now accepted by cleanup detector |
| facts memory status | Checks facts-memory file health | Direct local status check |
Core Modes
Wake, sleep, shutdown, command mode entry/exit, time, and Robie identity Shutdown halts Linux after goodbye/spin sequence
shutdown command triggers a goodbye/spin sequence and then halts the Linux OS. Keep physical access nearby when testing this command.Local greeting
AI mode
Goodbye sequence / sleep
AI / command mode
Goodbye/spin sequence, then halts Linux
AI / local handler
Any phrase containing control mode, command mode, or man mode
Enters movement command mode
AI / local handler
enter commando and commando mode to command modeLeaves command mode
Command mode
Speaks clock-style time response
AI mode
Gives fuller Robie identity/personality description
AI / local handler
what are you doing misrouteConversation & Curiosity
Occasional conversational and visual follow-ups, explicit Interview Mode, and permission-based idle questions Conversational, visual, and interview features work for everyone; idle curiosity is owner-only
No activation phrase. This is normally enabled for every active speaker during ordinary AI conversation.
After fully answering, Robie may end with exactly one short question relevant to what the speaker just said. If the speaker gives a clear, meaningful answer, Robie may acknowledge it and ask exactly one more specific question about a detail in that answer.
AI conversation / every speaker
Describes visible facts and, when clearly supported, adds one cautious observation about what appears notable, useful, unusual, or relevant. He may then ask exactly one short question grounded in visible evidence.
Scene View, Object Inspect, and Object Examine / every speaker
Begins interviewing the current speaker, briefly acknowledges each response when appropriate, and asks exactly one concise next question per turn.
AI conversation / every speaker
Ends the current interview and returns to ordinary AI conversation.
Active Interview Mode
stop asking me questions ends that interview without turning off the normally enabled conversational-curiosity preference.Temporarily enables or disables conversational follow-ups, visual questions/awareness, and owner-only idle offers.
AI / runtime preference
interview me command can still start one.During an appropriate quiet conversational moment, Robie may ask permission before posing one bounded curiosity question. Guests do not receive owner-memory-based idle questions.
Accepts a clear yes or no and continues normally.
AI / authorized owner only
Speaker change, sleep, shutdown, Command Mode entry, Keep Watch start, or Walkabout start.
Clears Interview Mode, any pending deeper conversational probe, and any pending idle-curiosity permission request.
Automatic state protection
Status & Battery
Casual status, short status, full reports, facts memory status, environment, air analysis, main battery, and Bluetooth microphone battery Use how are you feeling for short status; status report for full report
how are you is casual. how are you feeling is a short status-style reply. status report or system report produces the long detailed report.Gives short casual banter/status reply
AI / local handler
how are you feelingGives short activity reply
AI / local handler
Gives short status-style health/environment summary
AI / local handler
Reports detailed system, camera, memory, USB audio, main battery, Bluetooth microphone connection and battery, and environment status
AI / local handler
Reports facts-memory file health
AI / local handler
Reports Robie's main battery and the Razer battery percentage when connected; otherwise reports that the Bluetooth microphone is off or disconnected
AI / local handler
Quick environmental-sensor snapshot
AI / local handler
Runs deeper on-demand air/VOC trend analysis
AI / local handler
Camera, Media & Sharing
Vision description, current visual reference, transcript sharing, child-style drawing, photo/video capture, email sending, cleanup Scene View stays separate from Object Inspect/Examine
what do you see?, Robie keeps the current visual reference file as his current visual reference until a new scene image replaces it, the service restarts, or the file is deleted. Follow-up questions can use the stored detailed observation or re-analyze the same image.email that, send that, email me that, send me that, and transcribe that refer to Robie's most recent completed AI answer. Phrases such as send that to me, email that to me, or send it to me remain camera-media commands.Emails a dated transcript containing the owner’s question and Robie's answer, with the exact completed Piper speech WAV attached
AI / owner-only local handler
First speaks one of several hold-still messages explaining that the task may take a minute or two. After that message finishes, he captures one new camera photo, submits it to GPT Image 2 with the installed child-drawing style reference, crops the result to an exact 3:2 frame that removes Robie's lower shell, adds Robie's transparent signature at bottom-right, and saves a borderless 3000×2000 JPEG tagged at 300 DPI. Following confirmed email delivery, he speaks one of several completion messages and asks what the recipient thinks.
AI / camera and image-generation task
Captures a fresh the current visual reference, describes it, and stores it as current visual reference
Scene View
Uses stored detailed visual context or re-analyzes the current visual reference file without taking a new photo
Scene View / current visual reference
Describes front camera view
Scene View
Saves timestamped photo locally
AI / local handler
Any send, email, or mail request that also includes photo, picture, image, or pic, such as take a photo and send it
Takes photo, emails it, then deletes local copy if send succeeds
AI / local handler
Records short video locally
AI / local handler
Any send, email, or mail request that also includes video or clip, such as record a short video and send it
Records video, emails it, then deletes local copy if send succeeds
AI / local handler
Sends newest saved photo/video
AI / local handler
Sends newest saved photo
AI / local handler
Sends newest saved video
AI / local handler
Sends newest camera media file
AI / local handler
Delete words delete, erase, clear, remove, clean, cleanup, purge, wipe, or trash plus camera/photo/video/media wording; examples: erase media, delete all photos, delete camera files, clear camera media, delete photos and videos, delete videos
Deletes saved camera photos/videos
AI / local handler
General wide-angle view and regular camera media functions
Scene View
Captures temporary object image and answers about the presented object
Object Inspect
Records temporary 3-second video, extracts frames, analyzes object across angles
Object Examine
Alright. Hold it steady and rotate it slowly.Continues discussion using recent object context
Object Inspect / Object Examine
Starts silent PIR-only room-monitoring mode
Keep Watch
Stops Watch Mode or cancels pending alert
Keep Watch
PIR-only trigger while Keep Watch is armed
Emails one event photo plus one 10-second event video
Keep Watch
Object Inspection
Look at this / Examine this / Analyze this object-helper commands Object Examine is now 3 seconds
Takes a temporary still image and analyzes the presented object
Object Inspect
Hold it steady.Records 3-second temporary video, extracts key frames, analyzes the object from multiple angles
Object Examine
Alright. Hold it steady and rotate it slowly.Uses recent object-inspection/examination context for follow-up answers
AI mode
Uses camera plus audio context to help inspect parts/markings
Object Inspect / Object Examine
Uses camera to identify/describe record sleeves, labels, or condition clues
Object Inspect / Object Examine
If a person is shown, describe only visible non-sensitive details
Avoids identity claims
Object Inspect / Object Examine
Command Movement
Manual forward/back/left/right, dynamic speed adjustment, stop, sensors, override Physical motor kill switch remains the real-world safety layer
Any command-mode phrase containing stop, halt, freeze, enough, or cancel; grammar also includes stop stop, please stop, stop it
Immediate stop
Command mode
Drives forward
Command mode
Drives backward
Command mode
Turns left
Command mode
Turns right
Command mode
Raises or lowers the currently active forward, backward, left, or right speed by 0.05 through a smooth motor ramp
Command Mode while moving
Disables command-mode collision sensors until cancelled or mode exit
Command mode
Re-enables command-mode collision sensors
Command mode
AI Movement & Gestures
AI movement, dance, show-off, spin, look-both-ways, duration modifiers Gestures are not navigation; AI movement now requires direct command wording
back up, go backward, go back a bit, move forward, turn left, or turn right. Conversational phrases such as I want to go back outside, go back to the old version, or move forward with the idea should remain conversation and should not move Robie.Any AI-mode phrase containing stop, halt, or freeze
Stops motion
AI mode
Dance/show-off gesture
AI mode
Spin/show-off gesture
AI mode
Look-both-ways gesture
AI mode
look around is checked as camera vision before AI motion in special parsingShort timed forward motion
AI mode
move forward with the idea or other non-physical wording.Short timed backward motion
AI mode
go back as an AI-mode trigger. Use go back a bit, go backward, or back up for intentional AI-mode backward movement. AI movement ignores collision sensors by design in this build.Short timed left turn
AI mode
Short timed right turn
AI mode
Phrases containing tiny, small, little, short, quick, slightly, bit, long, far, etc.
Adjusts AI movement duration
AI / local handler
_motion_duration()User response after Robie asks for dance feedback
Classifies feedback positive/negative/neutral
AI / local handler
Walkabout Modev5.4.0
Command-only roaming mode, obstacle redirect, wake-word-free stop Never starts automatically; say stop to end
stop without the wake word to end the session. The former 120-second automatic session cap has been removed.Starts commanded Walkabout Mode
AI / local handler
Wake-word-free stop listener accepts command-mode stop phrases: stop, halt, freeze, enough, cancel; grammar also includes please stop, stop it, stop stop
Stops Walkabout Mode
Walkabout Mode
Robie first. With the timed session cap removed, this is the normal way to end Walkabout.Front sensors trigger redirect; rear sensor protects backup
Stops, backs up if safe, turns gently, and tries another path
Walkabout Mode
WALKABOUT_MAX_FORWARD_SECONDS = 6.0Start confirmation; low-chance redirect comments; stop/complete line
Keeps Walkabout mostly quiet
Walkabout Mode
Attention & Watch
Attention Mode and PIR-only Keep Watch start/stop/disarm behaviour Keep Watch and Walkabout suspend Attention Mode while active
Enables quiet-gap camera centering/attention behaviour
AI / local handler
Disables attention scanning/centering
AI / local handler
Starts silent PIR-only Watch Mode after arming settle window
Keep Watch
Stops Watch Mode
Keep Watch
No spoken command; PIR-only trigger while Watch Mode is armed
Captures event photo + 10-second event video, emails both
Keep Watch
No spoken command; automatic while Watch Mode remains active
Sends repeat alert packages only after cooldown if PIR motion continues or recurs
Keep Watch
Owner Memory & Privacy
Robie can maintain owner-authorized conversational continuity and personal preferences.
Speaker Identity
Guest speaker switching and Robie spoken-name variants. Protected owner recovery and unit identity controls are omitted from this public edition.
Normalizes the variant to Robie in greetings, direct address, robot-name questions, and common courtesy phrases
AI / speech normalization
Sets non-owner speaker mode
AI / local handler
Gives fuller Robie description
AI / local handler
what are you doing activity replyMath & Feedback
Simple arithmetic and dance feedback handling Feedback only inside short dance-response window
Answers simple spoken arithmetic
AI / local handler
Aliases & Safety Notes
Vosk mishears, overlapping phrases, safety caveats Read before testing Walkabout, Watch, and camera modes
| Alias / Phrase | Maps to / Means | Notes |
|---|---|---|
| street | Movement | Safety-sensitive |
| enter commando, commando mode | Mode switch | |
| Ruby / Rubi / Roby / Robby / Robbie / Rudy | Robie spoken-name variant when used in direct-address or robot-identity context | Context-bound; ordinary references to real people remain unchanged |
| email that / send that / transcribe that | Email last AI answer transcript + WAV | Owner-only; distinct from send that to me camera-media routing |
| look around | AI mode | Camera/media-sensitive |
| erase media | Camera cleanup | Camera/media-sensitive |
| go for a walk / walkabout | Walkabout Mode | Safety-sensitive |
| motor kill switch | Hardware | Safety-sensitive |
| pause/restore attention | Attention mode | Camera/media-sensitive |
| no AI chat while watching | Keep Watch | Camera/media-sensitive |
| PIR-only | Keep Watch | Camera/media-sensitive |
| Scene View vs Object Inspect vs Object Examine vs Keep Watch | Camera modes | Camera/media-sensitive |
| hold steady and rotate slowly | Object Examine | Camera/media-sensitive |
| go back outside / go back to... | Conversation only in AI mode | v5.4.5: should not trigger backward movement |
| go back | Command-mode backward phrase; not a bare AI-mode trigger | In AI mode use go back a bit, go backward, or back up |
| latest_view.jpg | Current visual reference | Replaced by each fresh Scene View capture; not durable memory |