Deformation-aware contact-rich manipulation skills learning and compliant control

In this paper, we study a vision-based reactive adaptation method for contact-rich manipulation tasks, based on the compliant control and learning from demonstration. For contact-rich tasks, the compliant control method is essential, especially when interacting with a deformable object with unknown properties, such as pizza dough. Learning from demonstration (LfD) provides a solution for this challenging task. However, the generalisation capabilities of LfD for deformable object manipulation tasks are still a challenging and opening issue, especially for unknown and dynamic tasks. Therefore, in this work, we investigate the vision and force-based perception feedback to enhance the generalisation of the LfD outcomes. The computer vision algorithm was developed to recognise the shape of the object and calculate the deviation between the desired shape and the current shape. The deviation of shape adjusts the parameters of learned primitive skills encoded by Dynamic Movement Primitives (DMPs). We adopt the pizza dough rolling task as the typical case to evaluate the performance of the proposed method. The shape and thickness of the dough can be made to the desired shape and thickness.