TwinFlow @ Aalto Mechatronics Circus 9.4.2026

This work presents a few-shot object detection approach that uses vision foundation models to detect unseen industrial objects from only a few samples. It eliminates the need for retraining by separating object localization and identification through prototype-based matching.

This project focuses on improving material handling in factory intralogistics through three key ideas. First, a tracking system combines camera and UWB data to locate objects accurately. Second, a vision-based quality inspection system uses synthetic data to detect defects with limited real data. Third, an autonomous mobile robot is developed to detect a trolley, navigate to it, and attach for automated transport. Together, these solutions show how sensing, perception, and robotics can improve efficiency and automation in industrial environments.

The project presents a smart pallet system featuring UWB-based real-time tracking and efficient crane-operated handling, integrated directly with overhead cranes and forklifts.

The project presents a prototype in which a vision system detects the position and orientation of a sheet-metal blank, enabling a robot to track, pick, and reorient the blank from a moving conveyor.