AlohaMini

Dual-arm mobile robot with lift — built from scratch on LeRobot

Двурукий мобильный робот с лифтом — собран с нуля на LeRobot

---

Overview / О проекте

EN: AlohaMini is a compact dual-arm mobile robot — 2× SO-101 6-DOF arms, 3-wheel omni base, motorized lift, and 5 cameras — running on Raspberry Pi 5 with HuggingFace LeRobot for imitation learning.

RU: AlohaMini — компактный двурукий мобильный робот: 2× SO-101 (6 степеней свободы), голономная 3-колёсная база, лифт и 5 камер на Raspberry Pi 5. Управление с геймпада, веб-интерфейс, запись датасетов и обучение политик через LeRobot.

---

Screenshots / Скриншоты

Arms 3D (disarmed / выключен)	Arms 3D (teleoperated / управление)

Camera feeds / Камеры	Gamepad binding / Биндинг осей

---

Hardware / Железо

Component	Qty	Notes
Raspberry Pi 5 (8GB)	1	Robot host — lekiwi_host
Feetech STS3215 12V servo	16	All motors (arms + base + lift)
Waveshare Bus Servo Controller	3	Left arm / right arm / base
12V Li-ion battery	1	Base + lift power
12V→5V DC converter	1	Feeds RPi from battery
USB cameras	up to 5	top / front / rear / 2× wrist
4-inch omni wheels	3	Holonomic base

Motor IDs / ID моторов

Bus	Function	Motor IDs
/dev/ttyACM0	Base (wheels + lift)	8, 9, 10, 11
/dev/ttyACM1	Right arm	1–6
/dev/ttyACM2	Left arm	1–6

Port order can shift after reconnect — verify with motor scan. / Порядок ttyACM* может меняться при переподключении — проверяй сканом.

---

Repo layout / Структура репозитория

kiborg-alohamini/
├── AlohaMini/                    ← Hardware: CAD/STL, BOM, URDF simulation
├── lerobot_alohamini/            ← AlohaMini LeRobot fork (ОСНОВНОЙ софт)
│   └── examples/alohamini/
│       ├── controller_v3.py      ← Web pendant: cameras, 3D arms, map, AI
│       ├── ui_main.html          ← Web UI (Drive / Cameras / Arms 3D / Map / Record / AI)
│       ├── ui_settings.html      ← Gamepad & deadzone settings
│       ├── ui_arm_settings.html  ← Per-axis deadzone calibration
│       ├── robot_display.py      ← Pi face display (7" screen)
│       ├── ui_robot_face.html    ← Face UI: eyes, e-stop, CPU/RAM/disk
│       └── convert_recording.py  ← Convert recordings → LeRobot Parquet format
├── lerobot/                      ← Upstream HuggingFace LeRobot (reference)
├── pi-config/                    ← Pi config (systemd, calibration, boot, robot_src)
└── pipeline/                     ← Training pipeline scripts (record→train→infer)

---

Quick Start / Быстрый старт

1. Pi (robot host)

ssh pi@<pi-ip>
conda activate lerobot
cd ~/lerobot_alohamini
pip install -e ".[feetech]"

# Run manually (or use systemd services — see pi-config/systemd/)
python examples/alohamini/lekiwi_host.py

2. PC (web pendant / веб-пульт)

git clone https://github.com/teslaproduuction/kiborg-alohamini.git
cd kiborg-alohamini/lerobot_alohamini
pip install -e ".[feetech]"

# Set robot IP / Укажи IP малины
export ROBOT_IP=<pi-ip>         # Linux/Mac
$env:ROBOT_IP = "<pi-ip>"       # Windows PowerShell

python examples/alohamini/controller_v3.py

Открыть / Open http://localhost:8080

3. Pi face display (optional 7" screen)

# On Pi
python examples/alohamini/robot_display.py
# Open http://localhost:8090

---

Web Interface / Веб-интерфейс

Tab	EN	RU
Drive	Omni base + lift + camera preview	База, лифт, камера
Cameras	All 5 MJPEG streams	5 MJPEG-потоков с робота
Arms 3D	Three.js joints + sliders + presets	Three.js визуализация рук
Map / Routes	2D dead-reckoning, waypoints, routes	2D карта, точки, автопроезд
Record	Episode recording with timer	Запись эпизодов датасета
AI Control	Run trained policy over network	Инференс политики по сети

---

Communication / Протокол связи

PC  ──PUSH──►  tcp://<pi>:5555  ──PULL──►  lekiwi_host   (commands / команды, JSON)
PC  ◄──PULL──  tcp://<pi>:5556  ◄──PUSH──  lekiwi_host   (observations + base64 JPEG)

Both sockets use CONFLATE=1 — only latest message, no queue buildup.

---

Training Pipeline / Пайплайн обучения

1. Record (Recording tab)
   ↓
2. convert_recording.py  →  Parquet + mp4  (LeRobot format)
   ↓
3. lerobot train --policy.type=act  (GPU PC, ~1GB VRAM for ACT)
   ↓
4. AI Control tab  →  inference → robot actions

See pipeline/ for ready-to-run scripts.

---

Gamepad Support / Геймпады

Controller	Connection	Use
DualShock 4 / PS4	USB or Bluetooth	Arm teleoperation (6-DOF × 2)
RadioMaster Pocket	USB Joystick mode	Base driving + lift

Per-axis deadzone calibration at http://localhost:8080/arm-settings.

---

Environment Variables / Переменные окружения

Variable	Default	Description
ROBOT_IP	(required)	Pi IP address
ROBOT_USER	pi	SSH username
ROBOT_PASS	(unset)	SSH password — prefer key-based auth
CMD_PORT	5555	ZMQ command port
OBS_PORT	5556	ZMQ observation port

Рекомендуется / Recommended: SSH key auth — ssh-copy-id pi@<ip>, leave ROBOT_PASS unset.

---

Pi Configuration / Конфигурация Pi

pi-config/ contains:

• systemd/ — autostart services (alohamini-host, alohamini-cam, robot-display)
• calibration/ — joint calibration JSON
• boot/config.txt — fan curve, UART settings
• robot_src/ — deployed source (lekiwi.py, lekiwi_host.py, lift_axis.py)

# Install systemd services on Pi
sudo bash pi-config/systemd/install_services.sh

---

License

MIT — see LICENSE.

Based on HuggingFace LeRobot (Apache 2.0).