Project information

Tracking unseen objects in dynamic industrial robotic contexts, like distribution warehouses, presents challenges in handling rearrangements and tracking items through temporal gaps. The complexity rises when robots encounter objects beyond their training sets. To address these challenges, the research introduces synthetic and real-world datasets, and proposes a paradigm with an efficient transformer module, which outperformed recent methods in experiments. As Research Assistant I led the public release of the implementation. This included the refactoring, cleaning and testing of the codebase and rerunning experiments to ensure proper functionality of the implementation.