Figure 7(a) shows the path tracking curves of the RRDS in the passive and active direct switching training system. It can be observed that the RRDS can achieve stable path tracking training. Moreover, the controller in [18] could suppress the impact of the system offset on the tracking performance. Figure 7(b)-(d) shows the velocity tracking curves of the RRDS in the passive and active direct switching training process. It can be observed that the robot in [18] could achieve velocity tracking in the passive training stage. However, in the active training stage, the velocity exhibited a large deviation and could not adapt to the velocity of the rehabilitee. Thus, the velocity coordination could not be realized between the human and the robot, which threatens the safety of rehabilitee. This shows that the controller of [18] cannot directly switch from passive to active training and that it can only make the rehabilitee walk in the passive mode for rehabilitation exercises. The controller design method was 10 12 14 16 2 4 6 8 applied to the model proposed in [19] to illustrate the superiority of the proposed method for passive and active direct switching motions. The tracking curve of the manipulator is the same as that of the model proposed in this study, and the simulation results are as follows. Figure 8(a) shows the path tracking curves of the manipulator in the passive and active direct switching motions. It can be observed that the manipulator achieved stable tracking and that the controller could suppress uncertain motion environments. Figure 8(b) and (c) plots the angular 10 12 14 16 2 4 6 8 Passive Training 1.2 1.4 0.2 0.4 0.6 0.8 1 05 81015 Time (s) (c) Response velocity tracking curves of the joints and for the passive and active direct switching motions, respectively. It can be observed that the manipulator velocity was constrained within the specified range for passive motion. Moreover, for active motion, the manipulator had the capability of velocity decision. This shows that the designed controller can contribute to stable tracking in the manipulator for both passive and active direct switching motions and that it can improve the safety and intelligence of the system. 1.2 1.4 Active Training Direct Switch x-axis Position (m) (a) 0.2 0.4 0.6 0.8 1 2 46 810121416 05 81015 Time (s) (b) 0.2 Active Training 0.1 Passive Training Direct Switch -0.1 -0.2 0 58 10 Time (s) (d) Reference FIGURE 7. Path tracking of RRDS and velocity tracking curves of each motion axis [18]. (a) Path tracking. (b) Velocity tracking of the x-axis. (c) Velocity tracking of the y-axis. (d) Velocity tracking of the angle. 15 Active Training Passive Training Direct Switch Active Training Passive Training Direct Switch 1.2 1.4 Active Training Passive Training Active Training Direct Switch 2 468 10 12 14 16 q1 Position (m) (a) 0.2 0.4 0.6 0.8 1 58 Time (s) (b) Reference Response FIGURE 8. Path tracking of manipulator and velocity tracking of the double joint. (a) Path tracking. (b) Velocity tracking of q1. (c) Velocity tracking of q2. MARCH 2023 IEEE ROBOTICS & AUTOMATION MAGAZINE 81 10 15 1.2 1.4 Passive Training Direct Switch 0.2 0.4 0.6 0.8 1 58 Time (s) (c) 10 15 Active Training Passive Training Direct Switch q2 Position (m) q1 Velocity (m/s) y-axis Velocity (m/s) y-axis Position (m) q2 Velocity (m/s) Angular Velocity (rad/s) x-axis Velocity (m/s)