Team SWOT from THWS takes 2nd in RoboCup @Work

The Robocup @Work Challenge simulates a smart factory environment where robots must autonomously transport goods, recognize and manipulate objects, and navigate increasingly complex scenarios. Each day of the competition introduces new challenges—obstacles, decoys, and changing backgrounds—testing the limits of robotic perception and adaptability.

Building the Platform

Team SWOT’s robot was the result of years of development. Their platform featured:

• Omni-directional wheels for agile movement
• Dual laser scanners for forward and reverse navigation
• A UR5e robotic arm with a custom gripper and integrated camera
• A compact compute box containing a standard PC with GPU for real-time processing

All components were tightly integrated, with sensor and robot data converging on a central compute unit—ensuring fast, reliable decision-making.

Preparation and Machine Learning

Preparation began with localization—refining the robot’s awareness of its position in the arena. The team then used a synthetic data environment to train object recognition models. By generating tens of thousands of images with varied lighting and backgrounds, they created a robust machine learning framework that generalized well to real-world conditions. Their lab setup reflected the competition arena, allowing for iterative testing and refinement.

Their integration of the UR robot included leveraging the UR client library and a custom driver wrapper to achieve full control. The DC-DC control box proved especially useful, and the team emphasized the importance of reliable hardware, especially the manipulator with a force-torque sensor.

Advice for Future Competitors

Team SWOT shared valuable insights for aspiring teams:

• Invest in a reliable manipulator—it’s the heart of your system
• Object recognition takes real development—don’t underestimate it
• Hardware integration is time-consuming—plan ahead
• Build a diverse team—include lab engineers for continuity and integration expertise
• Keep it simple—fewer components mean fewer points of failure
• Simulate and test—know how your manipulator fits and functions before you arrive