Safety and Performance Co-Optimization in Human–Robot Interaction: An Adaptive Energy Tank Design With Dynamic Control Barrier Functions

Theco-optimization of safety and task performance is a central challenge in physical human–robot interaction. Passivity ensures safety by limiting system energy, and impedance control enables compliant execution. However, conventional energy-tank methods often rely on fixed power constraints and produce discontinuous control signals, severely limiting adaptability in dynamic environments. To overcome these limitations, this article introduces a novel safety-aware passive control framework that unifies high-order dynamic control barrier functions (HD-CBFs) with an energy tank. Our contributions are twofold: an HD-CBF-based energy tank that continuously enforces time-varying power bounds, and an adaptive iterative learning scheme that autonomously optimizes power profiles for complex tasks without compromising safety. Extensive experiments on a 7-DOF Franka Emika Panda manipulator validate the framework’s robustness across key scenarios, including task interruption, unstructured environmental variations, and clinically relevant robotic ultrasound scanning. The system consistently maintains real-time power constraints and reduces force-tracking root-mean-square error by 53.9% compared to baselines. This demonstrates an effective balance between safety and performance, making it highly suitable for real-world human–robot collaboration.