Bidirectional energy-flow modulation for passive visual admittance control using vision-force measurement fusion

Precision and compliance are essential to high-contact robotic tasks, while contact oscillations can be vital to both the end effector as well as the operation object. Passivity-based energy regulation of robot manipulators is a promising way to maintain decent operation behavior. However, current passive admittance methods in Cartesian-space depend on accurate pose reconstruction, making them difficult to apply directly to vision-guided high-precision alignment tasks. Furthermore, conventional passivity-based control usually requires large damping or strong energy constraints, leading to conservative control behavior. To address these limitations, a three-layer dynamic model consisting of nominal, offset and measured states is constructed, and a bidirectional energy-flow modulation mechanism is introduced to adaptively regulate energy exchange according to the real-time status. Moreover, a PID-based energy compensation, an adaptive virtual damping and an energy tank mechanism are integrated to constrain energy injection, dissipation, and release, thereby ensuring closed-loop passivity preservation while maintaining dynamic performances. Bolt alignment experiments are conducted under different initial positions, tilt-angle deviations and strong-contact conditions. At initial position 1, compared with shear thickening fluid (STF) visual admittance method, the proposed method reduces the peak contact force to 19.21N, shortens the alignment time to 6.11s, and decreases the residual contact force after alignment to 0.73N, corresponding to reductions of 65.4\%, 39.14\%, and 81.09\%. At the more challenging initial position 2, it still completes stable alignment with a minimum peak force of 5.58N and the shortest time of 6.04s. These results demonstrate improved impact suppression, dynamic response, task efficiency and reduced conservativeness.