PROJECT / 01ROBOTICS / MECHATRONICS / CONTROLS

6-DOF Robotic Arm + Digital Twin

Designed and built a 6-axis robotic arm with embedded motion control, computer vision, inverse kinematics, network communication, and a real-time Unity digital twin.

Mechanical DesignRoboticsEmbedded SystemsComputer VisionControlsUnityRaspberry PiESP32

HERO / 01Physical Object Detection

HERO / 02Unity Digital Twin

01 / OVERVIEW

One control stack connects the physical arm and digital twin.

A Raspberry Pi handles planning and vision, an ESP32 controls motion through a PCA9685 servo driver, and Unity visualizes live state, target poses, and predicted trajectories.

02 / SYSTEM ARCHITECTURE

From digital intent to physical motion.

The architecture separates visualization, perception, control, and actuation while keeping robot state connected across the system.

Unity Digital TwinVisualization · commands

WI-FI / TCP

Raspberry PiPlanning · vision · TCP

USB SERIAL

ESP32Motion control · USB serial

CONTROL / STATE

PCA9685Servo signal distribution

CONTROL / STATE

Servo MotorsJoint actuation

CONTROL / STATE

6-DOF Robot ArmPhysical system

03 / MECHANICAL DESIGN

Six-axis modular assembly.

CAD assembly of the six-axis robotic arm and gripper — 03 / 01CAD Assembly

Full six-axis robotic arm assembled on a desk beside its control electronics — 03 / 02Full Robot Arm

ACTUATORS

J1: Base; TD-8135MG
J2: Shoulder; DS3240
J3: Elbow; DS3240
J4: Wrist; DS3225
J5: Wrist; DS3218
J6: Wrist; DS3218
GRIP: Gripper; MG90D

Servo-driven joints connect 3D-printed PETG structures with an integrated gripper and camera mount.

04 / JOINT DEVELOPMENT

Joint and end-effector development.

Exploded CAD view of the robotic arm base joint assembly — 04 / 01J1 Joint Assembly

Underside CAD view of the robotic arm base joint assembly — 04 / 01J1 Joint Assembly

CAD turntable showing the end effector and eye-in-hand camera mount poster — 04 / 02End Effector + Eye-in-Hand Camera

Designed the joint interfaces and end effector around the selected servos, with an eye-in-hand camera mounted above the gripper for object detection.

05 / EDGE VISION

Custom-trained cube detection, deployed on the robot.

Raspberry Pi 4 with a connected Camera Module v2 — 05 / 01Raspberry Pi 4 + Camera Module v2

YOLO11n cube detector training and validation metrics across 100 epochs — 05 / 02Cube Detector Training Results

Trained a YOLO11n cube detector for 100 epochs on a custom 500-image dataset and exported it to NCNN for inference on a Raspberry Pi 4 with Camera Module v2.

06 / COMPUTER VISION

Connecting image coordinates to the robot workspace.

Object detection and pixel-to-table coordinate mapping demonstration poster — 06 / 01Object Detection + Coordinate Mapping

Fixed-observation table XY homography calibration dashboard — 06 / 02XY Homography Calibration

At a fixed observation pose, an XY homography maps detected cube locations from camera pixels to robot-table coordinates for motion planning.

07 / UNITY DIGITAL TWIN

Live state, previewed motion, and physical synchronization.

Unity digital twin interface showing the robotic arm and joint controls — 07 / 01Unity Digital Twin

Physical robot arm pickup demonstration shown alongside its Unity digital twin poster — 07 / 02Robot Arm Pickup Demonstration

Built a Unity digital twin for live robot-state visualization, ghost-pose target preview, and trajectory prediction synchronized with the physical arm.