Table 2.

Comparison of the different neural networks. Values display the difference in nRMSE values. Positive values show an improvement compared to the baseline (MLP or CNN from scratch).

		Joint Angles
		LSTM vs. MLP [%]	CNN vs. MLP [%]	Pretrained CNN vs. CNN [%]
	flex/ext	20.25	35.57	−5.79
hip	add/abd	7.47	13.43	−4.81
	int/ext rot	−25.20	2.80	5.40
	flex/ext	11.41	27.70	−6.38
knee	add/abd	−8.55	5.31	−9.50
	int/ext rot	−14.03	4.18	1.21
	plantar−dorsiflex	−2.38	12.53	−9.88
ankle	inv/ev	−5.61	2.87	−3.03
	int/ext rot	−7.67	2.61	−1.08
		Joint Moments
		LSTM vs. MLP [%]	CNN vs. MLP [%]	Pretrained CNN vs. CNN [%]
	flex/ext	4.13	20.64	−9.27
hip	add/abd	3.07	17.29	−1.83
	int/ext rot	−5.73	6.37	−2.07
	flex/ext	−10.49	5.59	−4.01
knee	add/abd	0.18	12.17	−5.76
	int/ext rot	−1.81	7.89	−2.39
	plantar−dorsiflex	7.30	13.90	−0.83
ankle	inv/ev	−7.39	−0.62	4.92
	int/ext rot	−4.79	3.37	−3.99