New LfH framework learns user-preferred safety interventions for haptic control

A new paper from Dawei Zhang and Roberto Tron introduces a Learning from Haptics (LfH) framework that adapts safety interventions in haptic human-robot shared control to individual user preferences. Traditional haptic guidance systems rely on predefined intervention strategies that cannot accommodate diverse user preferences or application scenarios. The LfH framework overcomes this by using a differentiable Control Barrier Function (CBF)-based optimization layer that automatically adjusts safety parameters to match demonstrated haptic responses. Instead of tuning controller parameters directly, users teach the system how they expect it to intervene during teleoperation. This results in haptic guidance that reflects personalized intervention preferences while preserving the intuitive interaction of haptic shared control.

Experimental validation in both simulation and on hardware demonstrates the framework's ability to learn personalized safety interventions from sparse user input, significantly reducing the mismatch between generated haptic feedback and demonstrated preferences. The approach eliminates manual trial-and-error design, making it scalable for diverse applications like surgical robotics, autonomous vehicles, and industrial teleoperation. By enabling robots to intuitively understand individual safety preferences, this work brings human-robot collaboration closer to seamless, trust-based interaction.