Humanoids now walk, balance and reach with remarkable generality: one whole-body tracking policy follows references from a human, or from an end-to-end policy. That generality travels in the trajectory, and a trajectory alone carries limited information about the interaction it should produce: at contact, the executing controller determines how the robot behaves. Single-task policies usually reach hard interactions by optimising trajectory and controller together in simulation; general stacks usually assume a preset or hand-chosen controller. We present KPI, a promptable kernel for physical interaction between the trajectory source and an unmodified whole-body tracker. Instead of a controller fixed before the task, the trajectory source sends a contract: per direction, track, comply, or hold a force range. From tracking error and a wrench estimate, the kernel adapts the arms' stiffness, damping, reference and feedforward toward it at contact rate. We demonstrate KPI through an agentic framework: from one instruction, a vision-language agent writes both the reference trajectory and the contract, with no task-specific code. We demonstrate instruction-driven winch operation, door opening, and box transport, alongside scripted surface-interaction experiments. In the winch demonstration, the humanoid is able to turn a crank to hoist a second robot fully off the ground.
Figures & tables
Fig. 2: Agentic execution through KPI. (a) Analyzer writes the motion and the contract σ from Detector’s geometric evidence, and Verifier checks feasibility, stage outcomes and the remaining goal, returning rejected proposals for revision. Executor runs an accepted stage: a motion core supplies the nominal reference and KPI adapts the arm controller under σ from the interaction state. (b) One revision in the box task, numbered as in (a): Detector locates the two contact faces and the surrounding obstacles, Verifier rejects the original straight approach because it interferes with the tabletop, and Analyzer raises the hands before advancing while keeping both contacts.
Fig. 5: Winch operation. (a) Operation sequence. (b) KPI and KPI-fixed’s hand trajectories. (c) MCC ∗ failure case.
Fig. 6: Door opening and passage: grasp the handle, unlatch, step back, swing open, and walk through.
Fig. 7: Box transport and load robustness. (a) The five stages of transport and placement. (b) Maximum liftable load category. (c) SONIC fails to maintain multi-point contact with the box walls, preventing lifting.
Fig. 8: Board-writing results using different methods.