Active Haptic Exploration-Inspired Environment Identification and Force-Constrained Adaptive Impedance Control of Flexible-Joint Manipulators

Steady contact may provide insufficient excitation for environment identification, while probing consumes force margin, and flexible-joint dynamics limit command realization. Inspired by active haptic exploration, this study combines force-margin-gated finite probing, concurrent learning, double-buffer history refreshing, and adaptive impedance control. A measured history-rank condition terminates initial exploration. An actual approach-speed condition and a robust control barrier function address contact transition and sustained contact, with one-sided release limiting execution transients. Continuous learning gives exponential convergence for constant noise-free parameters with full-rank history; noisy, slowly varying estimation is ultimately bounded on fixed-stack intervals. In 27 matched simulations, active strategies reached the rank threshold in nominal and hard environments, whereas no probing did not. Median relative stiffness errors with gating were 3.35% and 3.64%, versus 33.22% and 53.20% without probing. Refreshing reduced slowly varying stiffness error from 17.80% to 4.62%. Their relative performance depended on the environment and parameter being estimated. The simulations demonstrate the value of constrained active information acquisition.