Samantha - My Current Humanoid Robot Project (Work in Progress)

Hello everyone!

I'd like to introduce Samantha, my current robot project. It's still under development, but I wanted to share the progress so far. Below I include a short video showing a few movement examples and some of the current voice commands.

The robot is built on an old powered wheelchair with a humanoid torso. It is a hybrid project that combines Synthiam ARC with several Python scripts (mainly for voice control). My goal has been to use the strengths of both systems rather than relying on a single approach.

Voice Control

I experimented with the Bing Speech Recognition skill, but it was too slow for my application and recognition wasn't always reliable. Modern smartphones, on the other hand, recognize speech extremely quickly and with excellent accuracy in most languages. Rather than fight those limitations, I decided to have the smartphone perform the speech recognition and send the recognized commands directly to ARC.

The result is voice control that feels almost instantaneous. At the moment I have only seven voice commands (shown in the video), but adding new commands is straightforward. I originally planned a multilingual system, but since I'm currently the only user, I simplified it to a single language.

Controllers and How They Connect

The robot currently uses:

• 2 EZ-B controllers
• 1 IoTiny controller

Connection model: all controllers are on the robot’s onboard WiFi network. ARC (on the ThinkPad) connects to those controllers using the Connect By Name Robot Skill.

Controller responsibilities:

• EZ-B #0 (Master): controls the arm servos
  • 20 servos total (10 per arm)
  • 5 joints plus a 5-finger hand per arm
• IoTiny (#1): controls head pan/tilt, LEDs, and audio functions
• EZ-B (sensors) (#3): dedicated to sensors

Sensor configuration (connected to the sensor EZ-B):

• RPLidar
• IC laser distance sensor
• Two Sharp IR sensors (ADC0 and ADC1)

The RPLidar replaced the Doppler radar and PIR sensor that were originally planned.

Network Configuration

I initially tried running all three controllers as WiFi Access Points, but that proved unreliable. The current setup is:

• EZ-B Master Access Point mode
• IoTiny Client mode
• EZ-B Sensors Client mode

The Connect By Name skill discovers all controllers automatically. Controller assignments are listed above.

Power System

The robot is powered by:

• 2 × 12 V 50 Ah batteries

Each servo group has its own independent DC/DC converter, completely separated from the controller power supplies. Every major subsystem has its own fuse. This lets me power individual sections independently during development and testing, which simplifies troubleshooting and improves safety.

Wheelchair Base

The base is an old Meyra electric wheelchair (1970s-1980s). Its electronics are simple:

• two 24 V motors
• one joystick
• relays
• power button

I considered replacing the original electronics with a Sabertooth motor controller, but decided to keep the original mechanics. Rather than controlling the motors electronically, three servos physically operate the wheelchair controls:

• one servo moves the joystick forward/backward
• one servo moves it left/right
• one servo presses the power button

Everything is controlled directly from ARC scripts, so no Xbox controller is required.

Head Design

The head moves in pan and tilt. The head model I found online was visually attractive but designed by an industrial designer rather than an engineer; the mouth is permanently attached to the face, with no room to install a lip servo. Instead of redesigning the head, I mounted a blue LED behind the mouth and use a microphone to listen to the speaker output. The LED lights whenever Samantha speaks, creating the illusion of moving lips. The eyes also contain LEDs that blink three times during startup.

Onboard Computer

A Lenovo ThinkPad is mounted directly on the robot. It’s larger than ideal (a mini PC would be cleaner), but during development I wanted a full keyboard and display directly on the robot for easier programming and debugging.

Power-Up Sequence

The robot uses multiple power switches so subsystems can be started in sequence:

1. EZ-B Master
2. IoTiny
3. Sensor EZ-B
4. ThinkPad power
5. ThinkPad system
6. Head pan/tilt servos
7. Left arm
8. Right arm
9. Wheelchair joystick power

Before any servos receive power, an initialization script automatically loads all servo minimum and maximum positions and places every servo into a safe resting position to prevent unexpected movement during startup.

ARC Robot Skills and Scripts

Current ARC Robot Skills:

• Connect By Name
• Auto Position
• SoundBoard V4
• Speech Settings
• TCP Script Server Raw
• RPLidar
• NMS Data Grabber

Custom scripts include:

• Init
• Shutdown
• Power On

The RPLidar skill is installed but still needs configuration. I also need to configure my OAKD camera for person tracking. The standard EZRobot camera is excellent for many tasks, but reliably tracking people at 5-7 meters is beyond its capabilities, so I plan to use the OAKD for longer-range person tracking.

Python + ARC Integration

One of the parts I find most interesting is the TCP Script Server Raw skill. A Python program running on the ThinkPad receives voice commands from a smartphone through a small HTML page. The Python script then sends JavaScript commands to ARC via the TCP Script Server Raw skill. This creates an extremely responsive voice-control system while keeping ARC as the central controller for hardware, sensors, and Robot Skills.

Next Steps

My next major milestones are:

• configuring the RPLidar
• configuring the OAKD camera for person tracking

Once those features are working, I’ll post another video demonstrating autonomous person tracking. Samantha is still very much a work in progress, but every month she gets a little closer to the robot I originally imagined.

A short video with voice control over smartphone: https://youtube.com/shorts/UAlhG9_nWm8

Here are 2 pictures: