Implemented
3D perception
A VL53L5CX Time-of-Flight sensor turns a depth grid into nine spatial directions.
- • 4 × 4 ranging at 4 Hz
- • Interrupt with 100 ms fallback
Back to Projects
personalOngoing
Meet Dum-E
Building an autonomous robotic companion around the ESP32-S3
Dum-E is my long-term personal robotics project: a small quadruped designed to perceive its surroundings, react with an expressive face, move organically, listen, speak, and remain maintainable through over-the-air updates. This page is the project hub and will evolve with each development milestone.
Read development phase 1Behavior preview
A face that reacts
This hardware footage shows the implemented behavior: the depth grid selects a direction, Brain turns it into a gaze command, and LVGL interpolates the pupils toward the target.
The phase 1 article details the sensor classification, the LVGL animation, and the FreeRTOS message queues.Target specification
Implemented
Expressive display
A 240 × 240 GC9A01 round display renders vector expressions and animated gaze.
- • LVGL 9.5
- • Two 240 × 40 SPI DMA buffers
Planned
Quadruped motion
Four MG90S servos, non-blocking interpolation, and organic movement sequences.
- • Dedicated 5 V power rail
- • 50 Hz motor task target
Planned
Voice interaction
I2S microphone and speaker paths for wake-word detection and audio feedback.
- • PSRAM-backed audio buffers
- • On-device keyword spotting
Planned
Wireless maintenance
Dual application slots make room for safe OTA firmware updates.
- • Flash layout ready
- • Two 4 MB OTA slots
- • Runtime integration planned
System architecture
Peripheral ICs transmit events via FreeRTOS queues. The Brain receives these events and forwards the commands.
Sensing
VL53L5CX - detection zone via I2C
Inertial measurement unit
LSM6DSOX - Tilt and movement via I2C
Microphone
INMP441 - Audio acquisition via I2S
Sensor event queue
Brain
Central thread for control
Command event queue
Display LCD
GC9A01 - With LVGL via SPI (DMA)
Motor kinematics
MG90 - Movements via PWM
Audio Out
I2S - Speaker
Network & OTA
Wi-Fi - Dual firmware slots
● Current firmware
What already runs
The firmware separates sensing, behavior, and rendering. Brain_thread is the sole owner of behavior state, and only the LVGL task manipulates the display.
Perception
VL53L5CX driver, interrupt-driven measurements, and spatial classification.
Behavior
A deterministic event-driven controller selects gaze and facial sequences.
Rendering
Vector poses are interpolated and flushed to the GC9A01 in six DMA-backed strips.
What comes next
The same message-queue model is designed to integrate motor commands, audio, networking, and OTA without letting peripheral timing leak into the central behavior logic.
Motion
A dedicated thread for servo commands.
Audio
One thread for wake-word detection and another for audio output, connected through interrupts.
Autonomy
A final state machine coordinating perception, expression, motion, audio, and wireless maintenance.
Development milestones
Phase 1 — August 2026
3D vision and expressive face
Coupling a Time-of-Flight sensor with a vector face rendered on an LCD.
Open milestonePhase 2 — planned
Kinematics and power management
Powering four servos safely and generating smooth, non-blocking quadruped motion.
Phase 3 — planned
I2S audio and wake word
Capturing microphone input and managing the speaker, processor load, and PSRAM memory.
Phase 4 — planned
State machine and OTA
Bringing every subsystem together and updating the robot without freezing its behavior.