Autonomous pick-and-place on an igus ReBeL cobot

Context

First-year Master's project in MESI, in a team of four (Daniel, Theresia, Arnaud, Hashem). The goal was pick-and-place with an igus ReBeL cobot without programming a trajectory. Its community ROS 2 driver was incomplete and its proprietary protocol (CRI) poorly documented.

What I did

• Reverse engineering of the CRI protocol, real-time heartbeat included.
• A complete ROS 2 pipeline. YOLO11n and RealSense locate, MoveIt 2 plans, the SCHUNK gripper grasps.
• Kinematics validated in MATLAB, then coded as a three-layer Python library.
• Project management (schedule, Gantt, budget) and a Norcan cart, built together with Hashem.

What I noticed and learned

• The vision loop recomputes the target at every cycle, so the arm's repeatability (about ±12 mm) stops being critical.
• Set validation criteria for each subsystem before testing.

My mistakes

• We machined the profiles by hand, three unplanned days. Outsourcing would have cost a few tens of euros more.
• The schedule slipped from May to June 2026 and the HMI remained a prototype.

Results

• A repeatable detect, plan, grasp, place cycle with no human intervention.
• Inverse kinematics solved for 100% of the scenario poses.
• 60% success over 20 trials, each failure traced.
• The first known open source ROS 2 library for the SCHUNK EGP 25 gripper on igus.

Project cost

All the equipment was provided by the Hall de Technologie of the University of Strasbourg.

Tech stack