New Impedance MPC Cuts Knee Exo Offset to 0.1 mrad vs 515 mrad Classical
A 500 Hz Kalman MPC achieves zero steady-state error under 15 Nm spasms.
Cao and Tang propose a new control approach for knee rehabilitation exoskeletons that balances trajectory tracking with safety during spasms or voluntary effort. Their Impedance Model Predictive Control (MPC) uses an algebraic feedforward to simplify knee dynamics into a double integrator, then solves a quadratic program with hard constraints on range of motion, torque, and velocity per ISO 13482. A Kalman disturbance state, driven by direct series-elastic actuator (SEA) spring deflection sensing (an intrinsic torque estimate, no EMG), provides offset-free tracking and sensorless Assist-as-Needed. The constant state matrix allows offline precomputation of the QP cost inverse, enabling 500 Hz operation with a multi-step horizon.
In benchmarks against seven controllers (sinusoidal tracking, isometric hold), the Kalman MPC achieved 0.1 mrad RMS, 0.1 mrad steady-state, and 0.2 mrad peak error under a 15 Nm spasm. Classical impedance at the same stiffness showed 515 mrad steady-state offset. Without the estimator, MPC still achieved 4.8 mrad RMS and 8.3 mrad steady-state. All MPC variants met the 87 mrad clinical criterion; no classical controller did. The architecture is designed for the 20-DOF MyoSuite myoLeg via coupling-aware per-joint QPs.
- Kalman MPC achieves 500 Hz operation with offline precomputed QP cost inverse.
- Steady-state offset of 0.1 mrad vs 515 mrad for classical impedance under same stiffness.
- All MPC variants meet the clinical criterion of 87 mrad; no classical controller succeeds.
Why It Matters
This control framework enables safer, more responsive knee exoskeletons for rehabilitation, handling spasms without compromising precision.