Closed-Loop Control of Robotic Cutting through Tactile Force Estimation

The properties of human skin, such as softness, viscosity, and friction, enable precise handling and manipulation of tools for various tasks. One particularly challenging task in robotic manipulation is operating a knife, where most approaches simplify the problem by rigidly attaching the knife to the robot's end-effector. In this work, inspired by the adaptive nature of human skin, we designed a gripper equipped with four tactile sensors embedded in soft silicone pads to predict the torque applied at the knife. This information is used for closed-loop control during a chopping task. Furthermore, we developed a search algorithm to identify the contact point between the knife and the object, facilitating continuous force estimation throughout the cutting process with an LSTM-based model. Our approach introduces tactile sensing as a novel feedback mechanism for robotic chopping, setting it apart from existing methods that rely on adaptive controllers or vision-based object detection algorithms. The integration of force feed-back, informed by tactile sensors, allows for real-time trajectory and force adjustments, substantially improving the efficiency and precision of the chopping process. Experimental results show that the proposed LSTM-based model can effectively map non-linear and noisy tactile data to estimate contact normal forces, making it highly suitable for closed-loop control in dynamic tasks like cutting.