Table 4.

Comparison of Human-exoskeleton coordinated control strategy.

Paper	Control approach	Advantages	Challenges
Amiri et al. (2019)	IMRAC	The comparison between the results of the IMRAC and conventional MRAC, shows the IMRAC converged faster and consumed less computational time than MRAC.	Lack of testing under disturbing conditions or in different environments.
Liang et al. (2018)	Based on LSTM.	Comparison with statistical regression methods based on LSTM, is stable, disturbed only by missing data and outliers, and has better inter-individual adaption.	Adaptability to a large number of subjects was not tested.
Zhang et al. (2019)	Based on hierarchical Lyapunov.	Compared with PD control, the moment of human-computer interaction can be minimized by setting appropriate controls.	There will be a delay in lower layer mode switching.
Chen et al. (2022)	Based on NFO.	It has faster calculation speed and recognition accuracy than GA and PSO algorithms.	The stability is easily affected by the parameters of the PD controller.
Ma et al. (2020)	Based on DMPs.	Trajectory features are extracted from sample trajectories, and new trajectories are constructed.	It is impossible to construct a general model, and information such as the subject's height and weight needs to be considered.
Guo et al. (2020)	Based on Lw-CNN.	Compared with SVM, the intent recognition speed is faster, and the real-time control is better.	The recognition accuracy is affected by the window size, and the accuracy decreases significantly when the sliding rate decreases.
Hua et al. (2019)	GA_PSO algorithm.	Compared with the sensitivity amplification control (SAC), the human-robot interaction is reduced by 0.6%, which reduces the burden of human-robot collaborative motion.	Gait conversion and label categories are not subdivided, and the recognition conversion efficiency is not high.
Du et al. (2018)	Variable impedance controller and SMILC.	Compared with PD iterative learning control algorithm (PDILC), the tracking error of SMILC is more minor, and the control system is more stable.	SMILC is prone to jitter when the state trajectory reaches the sliding mode surface.