Demonstration-Calibrated Port-Hamiltonian Retuning for Manipulation Policies

PHRetune is an offline method that derives controller gains for a frozen policy without evaluation rollouts or gain search. It learns a port-Hamiltonian model from demonstrations to estimate effort and energy associated with the policy's predicted actions, adjusting the downstream controller while preserving the policy and its action representation. PHRetune improves Diffusion Policy success by up to 9.4 percentage points across LIBERO suites and outperforms alternative methods on real-world tasks.