Multi modal grounding refers to the task of linking concepts between different modalities such as language and manual action concepts. Learning concepts across different modalities is of particular importance for robotic systems where data often has little intrinsic structure and suffers from potentially inaccurate sensor readings.