With the advancement of modern medical technology, robot technology has become increasingly pivotal in rehabilitation, particularly for stroke and spinal cord injury patients. Traditional body weight support (BWS) systems often suffer from limitations such as large footprints, unstable weight support, and inability to facilitate natural overground walking. To address these issues, we have designed a dynamic weight support following rehabilitation robot that provides precise, stable unloading forces and enables realistic ground-based gait training. This robot integrates mechanical innovation with advanced control strategies, leveraging robot technology to enhance rehabilitation outcomes. In this paper, we present the comprehensive design, control methodologies, and experimental validation of our system, emphasizing the role of robot technology in overcoming existing barriers.
The robot’s mechanical structure comprises a gantry framework, left and right cantilever lifting systems for weight support, and a differential wheeled chassis for mobility. This design ensures stability during operation while allowing compact dimensions suitable for clinical and home environments. Key parameters are summarized in Table 1. The gantry provides robust support, and the cantilever systems utilize ball screws driven by servo motors to adjust the unloading force dynamically. Pressure sensors feedback real-time force data, enabling constant force control. The differential wheeled chassis, with two driven wheels and four casters, offers high maneuverability and a small turning radius, essential for following patient movements smoothly. This integration of mechanical components underscores the advancements in robot technology for rehabilitation applications.
| Parameter | Value (mm) |
|---|---|
| Height | 1907.7 |
| Width | 960.0 |
| Length | 1053.0 |
| Lifting Range of Cantilever | 0-700 |
| Slider Movement Range | 0-400 |
The electrical system is centered around an industrial PC with an i5-8250U processor, which serves as the main controller. It processes data from pressure sensors and laser rangefinders via USB-RS485 communication and sends control commands to servo motors and drive wheels through CAN bus. A 48V30Ah lithium battery powers the system, with step-down modules providing 24V for sensors and the PC. This setup ensures low-latency communication and modular expandability within the ROS (Robot Operating System) framework, highlighting the efficiency of modern robot technology in real-time control applications. Key electrical components are listed in Table 2, demonstrating the integration of sensors and actuators that form the backbone of the robot’s functionality.
| Component | Specification |
|---|---|
| Industrial PC | i5-8250U |
| Laser Rangefinder Sensor | BL-400NZ-485 |
| Pressure Sensor | 0-50 kg / 0-5 mA |
| Servo Motor for Weight Support | Kinco SMC60S-0020-30MBK-5DSU |
| Drive Wheel for Mobility | Hollysys DS10240C |
| Power Supply | 48V30Ah Lithium Battery |
For dynamic weight support control, we analyzed the system dynamics to derive the relationship between motor rotation and the unloading force. The force provided by the left cantilever, for instance, is given by:
$$F_l = \frac{2\pi}{L} \left( T_M – T_P – T_B \right)$$
where $T_M$ is the motor torque, $T_P$ is the preload torque, $T_B$ is the bearing friction torque, and $L$ is the lead of the ball screw. The motor dynamics are described by:
$$J_M \ddot{\theta} + B_M \dot{\theta} = T_M$$
leading to the expression for motor torque:
$$T_M = \left( M_M \ddot{\theta} + B_M \dot{\theta} \right) \left( \frac{L}{2\pi} \right)^2$$
Thus, the unloading force relates to the motor angle $\theta(t)$ as:
$$F_l = Q(\theta)$$
where $Q$ incorporates dynamic terms. The total compensatory force $F_c$ is the sum of left and right forces:
$$F_c = F_l + F_r$$
We designed a PID-based control strategy to maintain constant unloading force. The error $e(t)$ between the desired force $F$ and measured force $F’$ is computed, and the position compensation $\Delta \theta$ is derived as:
$$\Delta \theta = K_p e(t) + K_i \int e(t) dt + K_d \frac{de(t)}{dt}$$
This compensation adjusts the servo motor position via the drive system, ensuring stable force output. For hemiplegic patients, we implemented an unbalanced weight support strategy, tailoring forces individually to each side. This approach leverages robot technology to provide personalized rehabilitation, addressing asymmetric impairments effectively.
Human motion perception and following control are critical for natural gait training. Using laser rangefinders mounted on the cantilevers, we measure the positions of the patient’s shoulders relative to the robot. The relative distance $l$ and orientation $\theta_H$ are calculated as:
$$l = \frac{l_{HL} + l_{HR}}{2}$$
$$\theta_H = \arctan\left( \frac{l_{HL} – l_{HR}}{d} \right)$$
where $l_{HL}$ and $l_{HR}$ are left and right shoulder distances, and $d$ is the sensor separation. The human position in the robot frame is updated as:
$$x_H = x_R + l \cos(\theta_H)$$
$$y_H = y_R + l \sin(\theta_H)$$
The desired robot state $p_d = [x_M^d, y_M^d, \theta_M^d]^T$ is set based on the human position, and the tracking error $p_{et} = p_d – p$ is transformed into the robot’s coordinate frame using a rotation matrix $T$:
$$\begin{bmatrix} e_x \\ e_y \\ e_\theta \end{bmatrix} = T p_{et} = \begin{bmatrix} \cos\theta_M & \sin\theta_M & 0 \\ -\sin\theta_M & \cos\theta_M & 0 \\ 0 & 0 & 1 \end{bmatrix} \begin{bmatrix} x_M^d – x_M \\ y_M^d – y_M \\ \theta_M^d – \theta_M \end{bmatrix}$$
A velocity tracking controller then adjusts the robot’s linear and angular velocities:
$$v_R = v_M^d \cos e_\theta + k_x e_x$$
$$\omega_R = \omega_M^d + k_y v_M^d e_y + k_\theta e_\theta$$
where $k_x$, $k_y$, and $k_\theta$ are control gains. The inverse kinematics for the differential drive chassis yield wheel velocities:
$$v_r = v_R + \omega_R d$$
$$v_l = v_R – \omega_R d$$
with $d$ as the axial distance between wheels. This control framework enables precise following, showcasing the adaptability of robot technology in responsive human-robot interaction.
To validate our design, we conducted experiments involving a healthy subject performing treadmill walking at speeds of 1.2 km/h, 1.8 km/h, and 2.4 km/h, with unloading forces of 10 kg, 20 kg, and 30 kg (distributed as 5 kg, 10 kg, and 15 kg per side). The force sensors recorded real-time data, and the PID controller maintained force stability. Results indicated that higher speeds increased force fluctuations due to dynamic responses, but the system kept errors within acceptable limits. The root mean square error (RMSE) and extreme values are summarized in Table 3, demonstrating the efficacy of our control approach in robot technology applications.
| Unloading (kg) | Speed (km/h) | RMSE | Max (kg) | Min (kg) |
|---|---|---|---|---|
| 5 | 1.2 | 0.2497 | 5.679 | 4.360 |
| 1.8 | 0.2792 | 5.932 | 4.486 | |
| 2.4 | 0.3210 | 5.914 | 4.329 | |
| 10 | 1.2 | 0.4256 | 11.01 | 9.119 |
| 1.8 | 0.4137 | 11.43 | 9.005 | |
| 2.4 | 0.4404 | 11.05 | 8.927 | |
| 15 | 1.2 | 0.4554 | 16.06 | 13.90 |
| 1.8 | 0.4866 | 16.20 | 13.78 | |
| 2.4 | 0.6511 | 16.30 | 13.38 |
In following control experiments, the robot tracked the subject’s straight-line walking at low (1.2 km/h), medium (1.8 km/h), and high (2.4 km/h) speeds. The initial and final distance errors, along with average tracking errors, are listed in Table 4. At higher speeds, tracking accuracy slightly decreased, but the robot maintained stable following, with errors remaining within practical bounds. This performance underscores the reliability of robot technology in dynamic environments, ensuring patient safety during rehabilitation sessions.
| Speed Category | Initial Distance Error (m) | Final Distance Error (m) | Average Tracking Error (m) |
|---|---|---|---|
| Low | 0.42 | 0.12 | 0.11 |
| Medium | 0.42 | 0.21 | 0.16 |
| High | 0.42 | 0.32 | 0.25 |
In conclusion, our dynamic weight support following rehabilitation robot successfully addresses the limitations of traditional BWS systems by integrating advanced mechanical design and control strategies rooted in robot technology. The system provides precise, stable unloading forces and enables natural overground walking, enhancing rehabilitation efficacy. Experimental results confirm the robot’s ability to maintain constant force and accurate tracking across various conditions. Future work will focus on optimizing structural components, improving control responsiveness, and incorporating artificial intelligence for adaptive training. Expanding clinical trials will further validate the system, driving innovations in robot technology for rehabilitation. This research highlights the transformative potential of robot technology in creating personalized, efficient, and safe rehabilitation solutions for patients with mobility impairments.