Dobot Magician Lite — YOLO Object Detection + Palletizing

RAS 545 · Robotic and Autonomous Systems Lab 3 · Arizona State University · Fall 2025

Autonomous perception-driven palletizing: YOLOv8 classifies objects in real time via a mounted camera, and the Dobot Magician Lite picks and sorts them into the correct pallet using a suction cup.

---

Demo

---

Overview

The Dobot Magician Lite (Serial: DT15-2311-1432) uses a camera mounted above the end-effector to detect objects placed in a fixed pickup zone. YOLOv8s classifies each object into one of two categories — food or vehicle — and the robot autonomously picks it up and places it in the corresponding pallet (A or B), then returns home to repeat the cycle.

Detection logic

YOLO Label	Category	Destination
apple, banana, sandwich, pizza, cake	Food	Pallet A
car, truck, bus, motorbike, bicycle	Vehicle	Pallet B

---

Repository Structure

.
├── camera_palletization.py          # Main YOLO + Dobot integration script
├── Lab_3_RAS_vishal.pdf             # Full lab report (IEEE/IJRR format)
└── README.md

---

System Architecture

Camera (cv2) → YOLOv8s Inference → Class Detection
                                         ↓
                              Food → Pallet A
                              Vehicle → Pallet B
                                         ↓
                         Dobot: Home → Pickup → Intermediate → Drop → Home

---

Technical Details

Robot States

State	X (mm)	Y (mm)	Z (mm)	R (°)
Home	240.000	0.000	150.000	−8.881
Pickup	306.420	−82.706	−55.166	−15.104
Intermediate	193.526	22.005	35.189	6.487
Pallet A (Food)	210.180	−233.553	24.075	−46.766
Pallet B (Vehicle)	303.931	220.869	26.487	36.006

Motion Sequence

Home → Pickup (descend) → suck(True) → Intermediate (lift)
     → Pallet A or B (drop) → suck(False) → Home → repeat

Key Parameters

Parameter	Value
Model	YOLOv8s (yolov8s.pt)
Camera index	/dev/video2
Resolution	640 × 480
Robot port	/dev/ttyACM0
Speed	100, 100
Home tolerance	±10 mm

---

Setup & Usage

Requirements

pip install pydobot2 ultralytics opencv-python
sudo apt install v4l-utils ffmpeg

Permissions

sudo usermod -a -G dialout $USER
v4l2-ctl --list-devices
ls /dev/tty*

Run

python camera_palletization.py

• The robot homes on startup, then waits at home between each cycle
• Live camera feed shows YOLO bounding boxes and detection status
• Press q to quit cleanly — the finally block closes the camera and serial connection

Update coordinates for your setup

home_coordinates         = [239.999, 0.0, 150.0, -8.881]
intermediate_coordinates = [193.526, 22.005, 35.189, 6.487]
pickup_coordinates       = [306.420, -82.706, -55.166, -15.104]
palleteA_coordinates     = [210.180, -233.553, 24.075, -46.766]
palleteB_coordinates     = [303.931, 220.869, 26.487, 36.006]

---

Lessons Learned

• Camera and robot coordinate calibration is critical — even a 5 mm offset causes consistent misses at the pickup zone
• Lighting significantly affects YOLO detection reliability; a uniform white surface gave the best contrast
• The is_home() check with ±10 mm tolerance prevents the robot from attempting a new pick cycle mid-motion
• Camera must be released (cap.release()) before robot motion and reopened after — USB bandwidth conflicts caused frame drops otherwise

---

Course Info

• Course: RAS 545 — Robotic and Autonomous Systems Lab
• Instructor: Prof. Sangram Redkar
• Lab Charge: Sai Srinivas Tatwik Meesala · Rajesh S Aouti
• University: Arizona State University, Tempe AZ
• Semester: Fall 2025
• Grade: 10 / 10