ROS 2 Medical Rehabilitation Robotics Guide 2026

Rehabilitation robots help patients recover movement after stroke, injury, or surgery — delivering repeatable, measurable therapy. This guide covers the ROS 2 stack for exoskeletons and therapy devices: compliant control, assist-as-needed algorithms, intent detection, and medical-grade safety.

1. The Rehabilitation Robot Stack

sudo apt-get install -y   ros-humble-ros2-control   ros-humble-ros2-controllers   ros-humble-realtime-tools

# Layers:
#   1. Actuator control  (torque/impedance, 1kHz)
#   2. Intent detection   (EMG / force / motion)
#   3. Assist controller   (assist-as-needed therapy)
#   4. Session logic        (exercises, progression, metrics)
#   5. Safety monitor        (ROM limits, force caps, e-stop)

2. Impedance Control

Rehab robots must be compliant — behaving like a tunable spring-damper around the patient's limb:

def impedance_torque(self, q, qd, q_desired):
    # tau = K (q_des - q) - D qd    (virtual spring-damper)
    # Low K early in therapy (patient leads), higher K later (more support)
    tau = self.K @ (q_desired - q) - self.D @ qd
    tau = np.clip(tau, -self.tau_max, self.tau_max)   # hard safety cap
    return tau
    # Adjusting K/D changes how much the robot 'helps' vs 'resists'.

3. Assist-As-Needed Therapy

The core principle: help only as much as the patient needs — encourage their own effort:

def assist_as_needed(self, tracking_error):
    # Deadband: within tolerance, provide NO assistance (patient works)
    if abs(tracking_error) < self.deadband:
        return 0.0
    # Outside the band, assistance grows with error, but capped
    beyond = abs(tracking_error) - self.deadband
    assist = self.gain * beyond
    # Slowly shrink assistance over sessions as the patient improves
    return np.sign(tracking_error) * min(assist, self.max_assist)

4. EMG-Based Intent Detection

Read muscle activation so the robot moves withthe patient's intention:

def emg_intent(self, emg_channels):
    # Rectify + smooth EMG -> activation envelope per muscle
    env = [lowpass(np.abs(ch)) for ch in emg_channels]
    flex = env[self.flexor] - env[self.extensor]
    # Threshold above resting noise = intended motion
    if abs(flex) > self.emg_threshold:
        direction = np.sign(flex)
        return direction * self.assist_velocity
    return 0.0   # trigger assistance the instant the patient tries to move

5. Gait Training

# Lower-limb exo replays a physiological gait trajectory, assisting
# each joint through the stance/swing cycle.
def gait_cycle(self, phase):        # phase 0..1 over one stride
    hip = self.gait_ref['hip'](phase)
    knee = self.gait_ref['knee'](phase)
    ankle = self.gait_ref['ankle'](phase)
    for joint, q_des in [('hip', hip), ('knee', knee), ('ankle', ankle)]:
        tau = self.impedance_torque_1d(joint, q_des)
        self.apply(joint, tau)
    # Body-weight support + treadmill sync complete the setup.

6. Progress Metrics

• Active participation: patient torque vs robot assist ratio — should rise over time.
• Range of motion: achieved ROM per joint per session.
• Smoothness: movement jerk metrics indicate motor recovery.
• Repetitions: high-dose repetition drives neuroplasticity.
• Trends: objective, quantified progress reports for clinicians.

7. Patient Safety Architecture

• Mechanical ROM stops: hardware limits inside the patient's safe range.
• Torque/force caps: can never exceed tissue-safe limits.
• Patient e-stop: the patient can stop the device themselves.
• Back-drivability: the device yields if the patient resists.
• Watchdog: any fault drops to a safe, supported hold.

8. Engagement & Gamification

• Serious games: map exercises to game goals for motivation.
• Real-time feedback: show effort and success to the patient.
• Adaptive difficulty: scale challenge to current ability.
• Tele-rehab: supervised home therapy with remote monitoring.

9. Regulatory Context

• IEC 60601: electrical safety for medical equipment contacting patients.
• IEC 62304: medical device software lifecycle.
• ISO 14971: risk management across the device.
• FDA / EU MDR: clearance before clinical/home use.
• ROS 2 typically powers research prototypes; products use certified stacks.

10. Platforms

• Lower-limb exos: gait training after stroke/spinal injury.
• Upper-limb devices: arm/hand therapy tables.
• End-effector robots: guide the hand through reaching tasks.
• Soft exosuits: lightweight, cable-driven assistance.

Key Takeaways

Rehabilitation robotics turns therapy into repeatable, measurable, motivating treatment. Impedance control makes the device compliant, assist-as-needed encourages the patient's own effort, and EMG intent detection lets the robot move with the patient's intention. Quantify progress through participation and ROM metrics, gamify for engagement, and build the whole system around patient-safe ROM limits, force caps, and back-drivability — under IEC 60601/62304 compliance before anyone is treated.