Table 2 Performance comparison of NARX-RNN and Kalman filter for real-time online prediction of FES-induced ankle joint torque with eEMG: root mean square errors (RMSEs), normalized root mean square errors (NRMSEs) and variance accounted for (VAF) are shown
From: Real-time estimation of FES-induced joint torque with evoked EMG
Patient/Subject | Estimator | RMSE (Nm) | NRMSE (%) | VAF (%) |
|---|---|---|---|---|
P1 | NARX-RNN | 2.13 | 6.08 | 92.23 |
| Â | Kalman filter | 6.27 | 17.91 | 83.05 |
P2 | NARX-RNN | 2.15 | 10.63 | 88.48 |
| Â | Kalman filter | 2.57 | 12.71 | 56.27 |
P3 | NARX-RNN | 0.24 | 21.24 | 78.75 |
| Â | Kalman filter | 0.31 | 27.31 | 75.16 |
H1 | NARX-RNN | 0.19 | 3.80 | 95.24 |
| Â | Kalman filter | 0.46 | 9.20 | 93.62 |
H2 | NARX-RNN | 1.28 | 10.50 | 77.68 |
| Â | Kalman filter | 1.97 | 15.74 | 69.29 |
H3 | NARX-RNN | 0.84 | 8.67 | 82.05 |
| Â | Kalman filter | 1.02 | 10.01 | 78.44 |
Average performance | NARX-RNN | 1.13 ±0.87 | 10.15 ±6.40 | 85.73 ±7.31 |
|  | Kalman filter | 2.10 ±2.22 | 15.48 ±6.67 | 75.97 ±12.65 |