Small scale · Intelligent motion

TORY

Physical AI Companion

When microrobots can sense and make decisions, even cramped, complex, hard-to-reach spaces become places we can explore.

Concept and design / Oscar Yeh · Oscar Digital Technology
Explore the project ↗← Back to main site
TORY four-corner helium floating companion concept
FLOATING COMPANIONConcept illustration
PROJECT STATUS

Project under revision

UNDER REVIEW

The project direction and development plan are being reviewed. The roadmap below is retained as a reference and will be updated as the plan evolves.

View the existing development roadmap

V0–V5 development concepts

V0Under review
Airframe and manual control

Helium lift, four-corner propulsion, tethered motion, and stop on link loss

  • Prototype development
  • Validation test
V1Under review
Altitude hold and obstacle avoidance

ToF / IMU altitude hold, collision avoidance, and manual takeover

  • Prototype development
  • Validation test
V1.5Under review
Follow the user

Person detection, follow distance, and latency

  • Prototype development
  • Validation test
V2Under review
Two-way voice interaction

Onboard and Bluetooth headset audio, latency, and echo

  • Prototype development
  • Validation test
V2.5Under review
Continuity of memory

Summary storage, preference retrieval, and response context

  • Prototype development
  • Validation test
V3Under review
Indoor navigation

AprilTag localization, route planning, and safe movement

  • Prototype development
  • Validation test
V4Under review
Return and charging

Low-battery return, alignment, and charging safeguards

  • Prototype development
  • Validation test
V5Under review
Full system integration

Voice, vision, memory, navigation, and sustained safe interaction

  • Prototype development
  • Validation test
01 / THE IDEA

Bringing movement and intelligence
to a new scale

TORY explores the intersection of AI microrobots and intelligent control. Sensors gather environmental data, a control system selects actions, and actuators move and perform tasks. Actual capabilities depend on size, materials, power, and surroundings.

01 — SENSE

Sense surroundings

Detect position, obstacles, or target signals to build useful environmental context.

02 — DECIDE

Decide

Adjust routes, speed, or behavior based on goals and live feedback.

03 — ACT

Act

Use scale-appropriate actuation to move, collaborate, or send data.

02 / POSSIBILITIES

Starting with research settings

Different scales call for different engineering solutions. These are areas to explore, not finished product features.

A / INSPECTION

Confined-space inspection

Explore pipes and equipment gaps that are hard for people to reach, gathering condition data.

B / ENVIRONMENT

Environmental monitoring

Measure changing conditions across a controlled setting for more detailed observations.

C / RESEARCH

Swarm collaboration research

Study how multiple units divide work, avoid obstacles, and share limited information.

03 / DEVELOPMENT PATH

Start with a verifiable prototype

Define the setting first, then choose size and actuation; validate control through small experiments.

“Micro” is not a fixed size. Millimeter-, centimeter-, and smaller-scale systems face different limits in energy, communication, manufacturing, and control.
01
Define task and setting

Specify the movement area, measurements, and success criteria.

02
Build a single-unit prototype

First verify movement, sensing, control, and data transfer.

03
Add intelligent control

Apply suitable models to localization, path planning, or anomaly detection.

04
Test collaboration and reliability

Compare task success and failure modes in repeatable test settings.

04 / FLOATING COMPANION

Integrated four-corner indoor floating AI partner

Built around low cost, validation, and iteration, Floating Companion uses helium for primary lift and motors for motion and attitude adjustments. The design aims to reduce power use and noise during sustained indoor flight, with protective and control measures for safer close interaction. This is currently a concept and prototype plan.

Concept illustration of the four-corner helium-powered AI Floating Companion
Helium envelope, four-corner propulsion frame, electronics, and proposed movement. Dimensions and performance remain to be validated.

Concept architecture

Lift and structureRounded rectangular helium envelope, ultralight carbon-fiber frame, and thin central electronics module
Propulsion and controlFour corner motors, ESP32-S3, motor drivers, and 1S LiPo
Connectivity and sensingPrimarily Wi-Fi; BLE for setup and nearby fallback; ToF and IMU planned for V1

Development stages / ROADMAP

V0 / AIRFRAME + MANUAL CONTROL

Current status: project under revision. This phase description is retained as an earlier concept, pending an updated development plan.

First verify safe lift and controlled movement

Concept flow for buoyancy platform, four-corner frame, ESP32-S3, manual control, and tethered tests
  1. Select a helium envelope and simulate electronics with dummy weights to approach neutral buoyancy.
  2. Build a four-corner propulsion frame using light carbon-fiber tubing and thin EPP or light sheet material.
  3. ESP32-S3 handles Wi-Fi, PWM, motor drivers, and link-loss protection; no Raspberry Pi or Jetson initially.
  4. Use a phone or desktop web UI to manually control forward/back, sideways, turning, fine altitude adjustment, and emergency stop.
  5. Use a safety tether for initial indoor tests of thrust, center of gravity, turning, and automatic stop on link loss.
V0 acceptance

No rapid rise or descent with motors off; controlled forward/back and lateral movement and turning; fine altitude adjustments; automatic stop on Wi-Fi loss; no motor-induced controller brownout or reboot.

V1 / HOVER + OBSTACLE SENSING

Altitude hold and basic obstacle avoidance

Concept diagram of ToF and IMU data controlling four motors via ESP32-S3
  1. Use a downward ToF sensor and PID control for altitude.
  2. Monitor pitch, roll, and yaw via IMU and correct attitude.
  3. Add front, left, and right ToF sensing for nearby collision avoidance and safe stops.
  4. Stop autonomous movement and enter safe mode at low battery.
  5. Users can take over or stop any automatic action immediately.
V1 acceptance

Maintain a set altitude; stop or correct near obstacles; sensor faults never drive motors to full power; users can take over at any time.

V1.5 / FOLLOW ME

Follow the user

Concept diagram of camera footage sent over Wi-Fi to a PC for detection and return follow commands

Send camera footage over Wi-Fi to an existing PC or server for person detection, then return high-level commands to ESP32. Goals: find the user, maintain distance, follow laterally, and face them.

V2 / AI VOICE + PERSONALITY

Voice and character interaction

Concept audio routing: onboard or Bluetooth headset microphone through a phone/PC gateway to onboard speaker or headset

Add a microphone, speaker, and AI voice via Wi-Fi to a PC or cloud AI, gradually developing real-time speech, an original character voice, personality, and contextual responses.

Two-way audio options / concept plan
  • Onboard mode: The onboard microphone captures speech; AI replies play through the onboard speaker.
  • Bluetooth headset mode: Pair the headset to a phone or PC; its microphone feeds AI via that device, which sends replies to the headset over Bluetooth.
  • Switching design: Choose input and output devices on the phone or PC. For onboard playback, send AI audio to the robot over Wi-Fi. Test call mode, latency, and echo cancellation based on actual device support.

ESP32-S3 Bluetooth remains for provisioning and control. It does not support Classic Bluetooth or LE Audio, so direct ESP32-S3-to-headset audio is not planned at this stage. Direct headset pairing would require different Bluetooth-audio-capable hardware.Espressif Bluetooth audio specifications ↗

V2.5 / MEMORY

Continuity of memory

Concept diagram of conversation summaries stored in SQLite for future context

Store important events, conversation summaries, and user preferences in a local SQLite database.

V3 / INDOOR NAVIGATION

Indoor navigation

Concept diagram of AprilTag localization, path planning, and safe movement

Start with low-cost AprilTag or room-marker localization; explore visual SLAM if needed. Goals include room recognition, finding the user, and simple path planning.

V4 / HOME BASE

Automatic return and charging

Concept diagram of return to base, alignment, and charging after low battery

Develop docking and charging so the robot returns to its home base, aligns, and charges on low battery.

V5 / FULL AI COMPANION

Full embodied AI partner

Concept diagram integrating voice, vision, memory, behavior, navigation, and safety control

Integrate voice, vision, memory, navigation, behavior, and safety control toward an embodied AI agent that can interact indoors for extended periods.

Connectivity and safety

COMMUNICATION

Connect control and AI over a home network

Primary communication uses 2.4 GHz Wi-Fi. For a multiroom space of roughly 165 m², use a router, access points, or mesh for coverage; BLE supports initial setup, maintenance, and nearby fallback.

Floating Companion → Wi-Fi → Home Router / Mesh → PC / AI Server、Smartphone、Internet / Cloud AI
SAFETY

Design in fail-safes from V0

  • Propeller guards, motors off by default at startup, and low-voltage protection
  • Stop on Wi-Fi command timeout or link loss, with emergency stop and manual takeover
  • AI issues only high-level behavior commands, never arbitrary raw PWM
  • Use safety tethers for V0 and V1 indoor tests
05 / FEEDBACK

Send feedback

Share your thoughts about TORY or Floating Companion.

Your feedback is saved to the site and visible only to the site owner.