Yongyan Cao
A whole-body impedance model predictive control method that allows floating-base robots to maintain balance and operate safely during physical human-robot interaction.
Floating-base robots (bipeds, humanoids) must maintain balance under rigid contact constraints and be robust to external forces during physical human-robot interaction (pHRI). Existing whole-body control (WBC) frameworks allocate the full joint space to locomotion or use fixed-gain impedance feedback, which accumulates steady-state error under sustained pHRI forces.
Proposes a three-layer architecture: (1) a centroidal model predictive control (MPC) plans contact forces over a 500 ms horizon; (2) a priority-driven WBC layer resolves balance into joint torques through contact-consistent null-space projection; (3) the residual null space is governed by a receding-horizon quadratic program (QP) that estimates and compensates for pHRI disturbances using a Kalman filter. Contact-consistent feedback linearization reduces the arm end-effector to a double integrator, enabling offline precomputation of the QP cost and operation above 1 kHz. A covariance-inflation protocol preserves disturbance estimates across contact-mode switches, and an Impedance Equivalence Theorem shows that the infinite-horizon limit recovers a classical task-space impedance law.
Simulations on a 17-DOF biped and the Unitree G1 humanoid validate the proposed method. Key contributions include the first whole-body impedance MPC framework for floating-base robots, robust disturbance estimation across contact-mode switches, and theoretical proof of impedance equivalence.