Towards intuitive and safe physical human-robot collaboration: an uninstrumented touchpad interface powered by intrinsic robot tactile sensing

Collaborative robots are increasingly prevalent in both industrial and domestic settings, demanding intuitive, safe and user-friendly interfaces for effective physical human-robot interaction (pHRI). External tactile sensors or buttons can enhance pHRI capabilities, but they often introduce additional complexity and cost. To address this challenge, we propose a novel framework that leverages the intrinsic tactile sensing capabilities derived from joint torque sensors integrated into commercially available collaborative robots. This framework powers an uninstrumented 3D-printed touchpad, mounted on the robot’s flange, where user touches are interpreted as virtual button commands by analyzing the resulting joint torque signals. Owing to their spatio-temporal nature, touch patterns are segmented using a windowing-based process and classified using deep learning models. A dynamic hierarchical finite state machine (HFSM) manages the interactive process, ensuring intuitive human-inthe- loop interaction and safety. A dedicated safety layer mitigates potential button touch misclassifications using safety validation protocols. A novel open dataset, pHRI-BTN-TACT, containing touch data from multiple participants, is proposed. Classification results validate the system’s ability to generalize, achieving userindependent performance with an online accuracy exceeding 93% across the robot workspace. The practical viability of the framework is demonstrated through three use cases on collaborative assembly. Demonstration videos and datasets athttps://ts-robotics.github.io/pHRI-BTN/.