To endow robots with autonomous behaviors in human populated environments, we require them to adapt to uncertain environments. This shift demands better perception, planning, and control systems, able to handle unpredictable interactions and noisy sensory feedback. In particular, being able to regulate the interaction forces becomes of paramount importance, ensuring precision or compliance according to the situation. Variable impedance control (VIC) offers a solution to the problem, regulating the stiffness based on perceived forces, robots can interact safely and avoid damage. However, manually coding these skills for every situation is impractical. The open question here is: which pieces of knowledge are needed for robots to encode these skills? Can the robot learn and re-use them in new situations?

In this direction, the research community presented different interesting approaches, which are based on exploratory behaviors or learning from human demonstration. In addition, recent AI methods, such as LLMs, demonstrated the capability to emulate human behaviors by means of natural interaction. Thus, why not exploit novel AI techniques to also regulate physical interactions in new contexts, by means of the combination of model-based and data-driven approaches? We believe that, enabled with such intelligence, robots can enforce safety and reliability, while building upon the principle of explainable AI. 

The aim of this workshop is to bring together passionate researchers with different expertise, such as control, learning and AI, to examine current research on how to successfully transfer compliance knowledge enabling natural behaviors, and establish future research directions.