URF: A Unified Robot Control-Policy Framework for Stable Contact Aware Manipulation

Researchers introduced URF, a unified robot manipulation framework that predicts a virtual target, stiffness, and an impedance-admittance switch ratio from multimodal inputs.

The controller-mode predictor is supervised with labels derived from measured contact forces, aligning the learned policy with real control behavior.

In box-flipping and line-pressing tests, URF outperformed admittance-only execution with better task success and fewer force spikes, oscillations, breakage, and safety stops.

The work tackles a major contact-rich robotics failure mode by improving the balance between tracking accuracy and safe physical interaction.