A robot that plans a trajectory and tracks it stiffly will break something the first time the world is not where the model said. Humans do not work this way: we modulate mechanical impedance and let physics absorb the discrepancy.

We build controllers on that principle — composing motion from primitives with well-defined energetic behavior, so safety comes from the structure of the controller rather than from a check bolted on afterward.

Current threads

• Physically consistent stiffness for spatial tasks
• Energy budgets as a safety guarantee during interaction
• Learning impedance policies without giving up stability

---

All research