A complete warehouse-automation line — conveyor, AI vision, a PLC-controlled robotic arm,
autonomous transport and live IoT monitoring — built for under 70,000 EGP.
Team SynTech · Faculty of Engineering, Benha University
Mechatronics & Automation Engineering · Supervisor: Dr. Amro Shafik
Siemens S7-1200 · TIA Portal / SCL · YOLOv8 · ROS 2 Jazzy · Nav2 · Flutter · Modbus TCP
Commercial warehouse automation starts at roughly $50,000 and runs into the millions. In Egypt, micro, small and medium enterprises are 90% of the private sector, 43% of GDP and 75% of the workforce — and almost none of them automate anything, because the entry ticket is out of reach.
SASS is a complete micro-fulfilment line that demonstrates the same architecture an industrial fulfilment centre uses — industrial PLC control, AI-based identification, autonomous transport and real-time monitoring — at a fraction of the cost, using components that can all be bought and serviced inside Egypt.
┌──────────────────────────────┐
│ Flutter mobile dashboard │ ≤ 2 s latency
└───────────────┬──────────────┘
│ Modbus TCP
┌──────────────┐ Modbus TCP ┌─┴──────────────────┐ PTO / PWM ┌─────────────┐
│ ESP32-CAM │ ─────────────► │ Siemens S7-1200 │ ────────────► │ SCARA arm │
│ + YOLOv8 │ SKU class │ CPU 1214C + SM1223│ │ 4-DOF │
└──────────────┘ (1–9) └─┬────────────────┬─┘ └─────────────┘
▲ │ │
│ HTTPS │ DI/DO │ Modbus TCP
┌──────┴───────┐ ┌───────┴──────┐ ┌─────┴──────────────┐
│ Cloud / FOMO │ │ Conveyor │ │ AGV — RPi 4B │
│ inference │ │ + IR sensor │ │ ROS 2 Jazzy / Nav2 │
└──────────────┘ └──────────────┘ └────────────────────┘
- The conveyor delivers an item to the pick-up zone and stops on IR detection (
< 2 ms). - The PLC triggers the vision node, which classifies the item into one of nine SKU categories.
- The classification is written into a Modbus TCP holding register as a single integer.
- The PLC solves the inverse kinematics analytically in SCL, on the PLC itself, and drives the arm.
- The vacuum end-effector places the item in its mapped 3×3 compartment.
- The bin counter increments and pushes to the dashboard.
- When a bin is full, the AGV is dispatched, delivers it, then returns or chains to the next full bin.
The AI never runs on the PLC. The vision node hands the controller one integer over Modbus TCP — which is what keeps the control layer deterministic and genuinely industrial.
 |
 |
| Conveyor — physically built. Aluminium-extrusion frame, PVC belt, pillow-block bearings, 12 V geared drive. | Full workcell CAD. Conveyor, 4-DOF SCARA arm, 3×3 storage cabinet and control panel. |
 |
 |
| AGV in simulation. Gazebo + RViz, navigating a mapped warehouse on a live 2D LiDAR scan. | URDF validated. Full TF tree verified in RViz before any hardware was touched. |
| Item | Specification |
|---|---|
| Reach / repeatability | 500 mm · ±1–2 mm |
| J1, J2 (base, elbow) | Closed-loop NEMA 17 (42BYGH40) + Leadshine CS-D508 drivers — encoder actually wired, so lost steps are detected |
| J3 (Z axis) | Pneumatic cylinder, Airtac MAL16 + 5/2 NAMUR solenoid valve, reed-switch end-of-stroke feedback |
| J4 (wrist) | MG996R servo, PWM from the PLC |
| End-effector | Vacuum — venturi ejector + 40 mm suction cup |
| Kinematics | Analytical inverse kinematics in SCL, executed on the PLC |
| Dynamics | Full Lagrangian model — M(q), C(q,q̇), G(q) — with per-axis PID design |
12 V DC geared motor with encoder (JGY-370) driven through a BTS7960 H-bridge, belt speed set directly from a PLC PWM output. An E18-D80NK IR reflective sensor detects arrival in under 2 ms and is wired straight to a PLC digital input. Stopping the belt is what triggers the vision capture — no polling loop, no race condition.
| Tier | Mode | Behaviour |
|---|---|---|
| 1 | Cloud YOLOv8 | ESP32-CAM sends the frame over HTTPS to a hosted endpoint and receives the SKU class. Highest accuracy. |
| 2 | On-device FOMO | On connectivity loss, a compact model trained on the same dataset runs on the ESP32-CAM itself. Fully offline, zero extra hardware cost. |
| 3 | Spare phone host | A spare Android phone hosts the full model on the local network as a manual last resort. |
The system does not depend on the internet to exist — only to be at its most accurate.
Raspberry Pi 4B running Ubuntu 24.04 with ROS 2 Jazzy, Nav2 and SLAM Toolbox, an RPLIDAR A1 for mapping and dynamic obstacle avoidance, and an ESP32-S3 co-processor holding a 100 Hz motor-PID and safety loop. Differential drive, ±5 cm target positioning, hardware E-stop independent of software.
Siemens S7-1200 CPU 1214C DC/DC/DC with an SM1223 16DI/16DO expansion module, programmed in TIA Portal V16 (SCL for kinematics, LAD for safety logic). Every node — PLC, vision node, AGV, tablet — sits on one private Wi-Fi LAN with a static IP. Modbus TCP is the primary protocol (native in TIA Portal, no licence, no extra hardware); MQTT via the Siemens LMQTT library is available as an optional path.
| Document | Contents |
|---|---|
| Power distribution & wiring (Rev. 8) | Full 24 V load budget, PSU sizing, fuse/MCB schedule, point-to-point wiring, complete PLC I/O map |
| Safety architecture | E-stop → dual-channel safety relay → contactor drop-out, PL d / Category 3 per EN ISO 13849-1 |
| SCARA dynamics & control | D-H parameters, Lagrangian derivation, mass/inertia and Coriolis terms, per-axis PID design |
| Vacuum gripper selection | Two-case force calculation, suction-cup and ejector sizing |
| AGV engineering spec | Chassis, drive kinematics, power budget, ROS 2 architecture, fault handling, BOM |
| Bill of materials | Every major component priced from real Egyptian suppliers |
| Subsystem | Status |
|---|---|
| Conveyor | ✅ Built and assembled |
| Workcell CAD | ✅ Complete |
| PLC control software | ✅ Modbus TCP client & server compiling, zero errors |
| AGV navigation | ✅ Running in Gazebo with LiDAR SLAM |
| Electrical design | ✅ Closed at Rev. 8 |
| Arm dynamics & control | ✅ Derived |
| Bill of materials | ✅ Fully sourced |
| Mechanical assembly & integration | 🔄 In progress |
Current stage: Prototype. Target for the fully integrated demonstration: 29 September 2026.
| Member | Role |
|---|---|
| Mohamed Abdeltawab Youssef | Team lead · Robotic arm & system integration |
| Mahmoud Mohamed Shamekh | CAD, kinematics, TIA Portal |
| Omar Mostafa Shokran | Computer vision · YOLOv8 |
| Youssef Mostafa Ayad | Conveyor electronics · PLC |
| Omar Mahmoud Metwally | AGV · ROS 2 & navigation |
| Marwan Ibrahim Zaki | AGV · Engineering documentation |
| Omar Atef | Mobile application · IoT dashboard |
Supervisor: Dr. Amro Shafik, PhD in Mechanical Engineering — autonomous systems, robotics, LiDAR technologies, mechatronic control and advanced mechanical design.
Automation shouldn't start at fifty thousand dollars.