Gen 7-Inspired AI Vision Cobot Arm: Camera-Guided Pick-and-Place with Human-Safe Collaboration

Factories sort small parts by eye and by hand thousands of times every day, and that work is slow, repetitive and easy to get wrong even when the parts look alike. Modern collaborative robots answer it with a camera above the workcell, a vision system that identifies each part, and an arm that shares the workspace with people instead of hiding behind a cage. Universal Robots' Gen 7 platform, set for its European debut at electronica 2026 in Munich (10 to 13 November 2026) in the Teradyne Robotics booth as reported on 4 November 2026, pushes that idea further with AI-ready arms, a new controller and teach pendant, and a tool flange that delivers data, power and safety at the wrist, demonstrated with AI-based bin picking and AI PCB inspection.

This student project translates that headline honestly to lab scale: a desktop servo arm, an overhead USB camera, an OpenCV colour and shape classifier as the explicit stand-in for neural bin picking, a magnetic tool mount as the stand-in for the Gen 7 tool flange, and a presence-sensor zone that pauses the arm when a hand enters, sorting coloured blocks and cylinders into bins.

The end-to-end mechanism runs from camera frame to sorted bin in five stages a student can trace and an examiner can test. The USB camera captures the workcell from above, and the Raspberry Pi Zero 2 W runs the OpenCV classifier, which thresholds each frame by colour in HSV space and reads each contour's shape, so every object becomes one row: colour, shape, position and target bin. The Pi sends the pick queue to the ESP32 over serial, and the ESP32 turns each target into joint angles and drives the five servos through the PCA9685 driver at the capped joint speed. A servo holds position because its internal potentiometer feeds the shaft angle back to a comparator that drives the motor until the error reaches zero, which is why a taught pose replays repeatably. While the arm moves, the presence sensor watches the safety zone, pausing motion the moment a hand enters, and the e-stop overrides everything until manually reset.

1. The builder assembles the arm: MG996R servos at the base, shoulder and elbow, SG90 servos at the wrist and gripper, all driven by the PCA9685 from the ESP32, with the Pi Zero 2 W, USB camera and presence sensor mounted over a fixed workcell tray.
2. The builder calibrates the vision: coloured blocks and cylinders are placed in the tray under normal room light, the HSV thresholds are tuned until each colour separates cleanly, and the shape rule is checked by showing one block and one cylinder and confirming the classifier names both correctly.
3. The builder teaches the arm: using the dashboard joint sliders, the arm is moved to the pick position over the tray and to each bin, the poses are saved, and a replay confirms the gripper returns to the same positions.
4. For the sorting demonstration, the builder loads eight mixed blocks and cylinders into the tray, starts the pick cycle, and watches the dashboard queue drain as each object lands in its matching colour bin.
5. For the safety demonstration, the builder moves a hand into the marked zone mid-cycle and shows the arm pausing, then withdraws the hand and shows motion resuming, and finally presses the e-stop and shows that motion stays halted until the button is manually reset.
6. For the tool-swap demonstration, the builder pulls the gripper off its magnetic mount, fits the pen-holder tool, and runs a taught drawing pose, showing that the wrist accepts more than one tool without rewiring.
7. The builder exports the cycle log as CSV, points to the pick order, the bin counts, the pause events and the e-stop event, and reports honestly in the report which sorts succeeded, which objects were misread, and under what lighting the demo ran.

Tech stack:

• ESP32 · motion controller and servo command host
• PCA9685 servo driver · 16-channel 12-bit PWM servo control over I2C
• MG996R servos · metal-gear servos for base shoulder and elbow
• SG90 micro servos · wrist rotation and gripper actuation
• Raspberry Pi Zero 2 W · vision computer running the classifier
• OpenCV colour and shape classifier · HSV thresholds plus contour shape
• USB camera · overhead workcell view for object detection
• Cobot cell dashboard · pick queue and bin counts and safety status

This cobot cell combines vision, motion and human-safety behaviour into one demonstrable loop, so the review story is a complete chain rather than one reading. The camera sees the workcell, the classifier names each object by colour and shape, the arm picks it and places it in the matching bin, and the safety zone watches for a hand and pauses the motion. The dashboard makes the whole loop visible: detected objects, the pick queue, bin counts, safety status, teach-mode joint sliders, a cycle log and a CSV export. The magnetic tool mount lets the same wrist carry a gripper or a pen holder, and the teach mode lets a student record and replay joint poses without writing new motion code. Every behaviour below is demonstrated in the same run, so the examiner never has to take a claim on trust.

• [Camera-guided pick-and-place] A USB camera mounted over the workcell feeds each frame to the classifier, which returns the colour, shape and position of every block and cylinder, and the arm picks objects in queue order and places each one in its matching colour bin — implemented in the base build.
• [OpenCV colour and shape classifier] The classifier uses HSV colour thresholds plus contour shape, counting four-vertex contours as blocks and near-circular contours as cylinders, and it is stated plainly as a colour and shape classifier, never as a neural network or a learned model — implemented in the base build.
• [Magnetic quick-swap tool mount] A magnetic plate at the wrist lets the operator swap the 2-finger gripper for a pen-holder tool by hand in seconds, echoing the Gen 7 idea of data, power and safety at the flange at student scale, with no electrical tool interface claimed — implemented in the base build.
• [Human-safe presence zone] An ultrasonic and IR presence sensor watches a marked zone around the arm, and the arm pauses mid-motion the moment a hand enters the zone and resumes only after the zone clears, so collaboration behaviour is demonstrable rather than asserted — implemented in the base build.
• [Physical e-stop and capped joint speed] A latching e-stop button halts all motion until it is manually reset, and joint speed is capped at 40 degrees per second in firmware, so the arm never moves at full servo speed near a person — implemented in the base build.
• [Teach-mode joint sliders] The dashboard exposes one slider per joint plus gripper open and close, and a taught pose can be saved and replayed, so a student can demonstrate teach-and-replay control without reprogramming the motion sequence — implemented in the base build.
• [Cobot cell dashboard with cycle log] The web dashboard shows the simulated camera view, the detected-object list, the pick queue, live bin counts, safety-zone status, joint sliders, a cycle log and a CSV export of every pick and place for the report — implemented in the base build.
• [Conveyor-fed sorting line] Future scope only, not implemented: feeding objects on a small conveyor instead of a fixed tray, listed as an extension rather than a claimed feature, since the base build sorts from one fixed workcell.

• Complete fabricated desktop arm (5 servo axes, aluminium brackets, 3D-printed joints, 2-finger gripper, pen-holder tool, magnetic tool mount, workcell tray, colour bins, coloured blocks and cylinders)
• Electronics set (ESP32 board, PCA9685 servo driver, MG996R and SG90 servos, Raspberry Pi Zero 2 W, USB camera, ultrasonic and IR presence sensors, latching e-stop button, 6 V servo supply)
• Vision and dashboard software (OpenCV colour and shape classifier, pick queue, cobot cell dashboard, teach mode, cycle log, CSV export)
• Setup and calibration guide (servo wiring, HSV threshold tuning, teach-pose procedure, safety-zone marking, e-stop test, CSV reading worksheet)
• Project report PDF (collaborative robotics background, Gen 7 comparison limits, vision pipeline design, safety design, demonstration results, honest measurement limits)
• PPT presentation for final review
• Viva Q&A preparation document (servo torque and payload limits, classifier vs neural model, safety-zone design, Gen 7 comparison limits, lighting dependence, single-workcell limits)