VTFLA Embodied AI Architecture

Why Baggage Handling Needs More Than Vision and Language

Standard bags vary in size, pose and surface; soft bags deform when gripped. A vision-language-action (VLA) model alone cannot feel a grip slipping. The VTFLA architecture extends the model with tactile and force channels - a multimodal embodied model that builds a perception-decision-execution closed loop for physical work:

• Vision - identify each bag, its pose and the scene layout
• Tactile & Force - control gripping force so bags are held stably without damage
• Language - understand task instructions and coordinate with the scheduling layer
• Action - execute gripping, transfer and stacking as one embodied behaviour

The result is stable gripping of complex baggage, flexible adaptation across robot forms, and a data loop in which operating experience continuously improves task strategy. VTFLA powers the airport baggage transfer solution and the robots behind it - DawnAnt and the wheeled dual-arm humanoid.