RoMaViD: Learning Robotic Manipulation from Video Demonstrations

Large-scale manufacturing necessitates automation, and robotic automation has emerged as a primary solution. Traditionally, robotic systems are designed for fixed assembly lines dedicated to specific product sets. However, with an increasing demand for specialized and customized products, there is a growing need for more agile manufacturing processes. To address this, we introduce a robotic assembly framework capable of generating assembly plans directly from RGB-D video demonstrations. We develop a pose estimation model to capture changes in object poses. We demonstrate its effectiveness in the robotic assembly of IKEA furniture, emphasizing its precision in managing assembly tasks.CCS CONCEPTS• Computing methodologies → Robotic planning; Vision for robotics; Activity recognition and understanding.