Last updated: October 21, 2026
Recovering from a stroke, a spinal cord injury, a torn ligament or major surgery depends on one thing more than any other: repeated, well-guided practice. Muscles and nerves relearn through hundreds or thousands of repetitions, and those repetitions have to be done correctly. The problem is that therapists’ time is limited, sessions are short, and most of the work happens at home, where nobody is watching.
That gap is where rehabilitation technology tries to help. Robots can guide a leg through thousands of steps, wearable sensors can count and measure movement, and smartphone cameras with motion-tracking software can check whether an exercise is done properly. Artificial intelligence is usually one layer within these tools rather than the whole product: it adjusts assistance, recognizes movement patterns or personalizes an exercise plan. The evidence for each of these is very different, and sometimes surprising.
Where Technology Fits in a Recovery Journey
Rehabilitation usually moves through stages, and each stage uses different tools. Seeing them in order helps explain why one device can be very useful early on and irrelevant later.
| Stage | Typical setting | Technology in use | What AI may add |
|---|---|---|---|
| Early recovery | Hospital or inpatient rehabilitation unit | Treadmill robots with body-weight support, end-effector gait trainers | Adapting how much the machine assists as the patient improves |
| Intensive therapy | Outpatient clinic | Overground exoskeletons, arm robots, virtual reality exercise systems | Detecting effort and fatigue, adjusting task difficulty |
| Return home | Home, by video or app | Wearable sensors, smartphone or tablet motion tracking, remote monitoring dashboards | Recognizing exercises, flagging poor form, alerting the therapist |
| Long-term mobility | Home and community | Personal exoskeletons, activity trackers | Gait analysis and fall-risk signals (mostly experimental) |
Walking Again After Stroke: The Strongest Evidence
The most robust evidence in this field comes from a Cochrane systematic review of randomized trials on electromechanical and robot-assisted gait training after stroke. These devices, either treadmill-based systems or footplate machines, move the legs through a walking pattern while supporting part of the body weight, allowing far more steps per session than a therapist could assist by hand.
Stroke patients walking independently at the end of treatment (per 100)
Illustrative estimates based on the Cochrane review of electromechanical-assisted walking training after stroke (2020), as summarized by PEDro; moderate-certainty evidence. Results vary by patient population, intervention and clinical setting; these figures should not be interpreted as guaranteed outcomes for individual patients.
The reviewers rated the evidence as moderate certainty and concluded that people who used these devices together with physiotherapy were probably more likely to walk independently. They also called for large pragmatic trials to establish the best frequency and duration, and how long the benefit lasts. The key word is together: the devices were tested as an addition to physiotherapy, not as a substitute. These trials evaluated electromechanical and robotic devices in general; they were not designed to measure the effect of AI features.
When More Technology Did Not Help: The Arm Robot Trial
Upper-limb recovery tells a more cautious story. The RATULS trial, which tested a robotic system rather than an AI intervention, was run at four UK centres and published in The Lancet in 2019, randomized 770 stroke survivors with moderate or severe arm weakness to one of three groups: robot-assisted training with the MIT-Manus robotic gym, an enhanced therapy program based on repetitive functional tasks, or usual NHS care. Both active programs were delivered for 45 minutes, three times a week, for 12 weeks.
Neither the robot nor the enhanced therapy improved arm function compared with usual care, and the enhanced human-led program did better than the robot on activities of daily living at three months. The authors concluded that their results did not support routine use of robot-assisted arm training as delivered in the trial, while acknowledging that some subgroups might still benefit.
Exoskeletons After Spinal Cord Injury
Powered exoskeletons are wearable frames with motors at the hips and knees. In 2014 the US Food and Drug Administration cleared the ReWalk Personal system for home and community use by people with paraplegia from spinal cord injury at levels T7 to L5, on condition that users train with the manufacturer’s program and walk with a trained companion. The Indego exoskeleton followed in 2016.
A 2016 meta-analysis of 14 studies covering 111 people with spinal cord injury found that, in supervised sessions after a training program, 76% of participants could walk with the exoskeleton without physical assistance from another person. In those studies, participants rated their effort at a level the authors compared with the perceived exertion of an able-bodied person walking at about 3 miles per hour (4.8 km/h). During training, falls occurred in 4.4% of participants, all with a first-generation device and none causing injury.
These are different outcomes and should not be confused: walking during a supervised session, safety during training, everyday use at home or in the community, and lasting clinical improvement over time. The first two are the best documented. Real-life use is more limited: in a Dutch study of 14 people with complete spinal cord injury who took a ReWalk home, participants used it on a median of 9 out of 16 days, mostly for exercise and social activities, and reported limits as an everyday mobility aid. Exoskeletons today are best understood as exercise and standing tools, not wheelchair replacements.
Home Programs With Motion Tracking
The fastest-growing area is home rehabilitation guided by sensors or a phone camera. These programs show the exercise, track the movement, give feedback on form and send data to a physiotherapist who adjusts the plan. This is where AI is used most directly, through motion recognition and pose estimation.
In a 2023 randomized trial in npj Digital Medicine, 140 adults with chronic low back pain were assigned to a fully remote digital program with motion-tracking sensors or to evidence-based in-person physiotherapy. Recovery in disability was similar in both groups, and fewer people dropped out of the digital program (15.7% vs 34.3%). The trial was conducted by researchers from Sword Health, the company that makes the program, so independent replication is important. It was a single trial in one condition, and it does not show that an app can replace a physiotherapist for other injuries or diseases. Products in this category also differ widely in what their software does, so the label “AI-powered” should be checked product by product.
Camera-based tracking has technical limits. A review of marker-free pose estimation for rehabilitation found that some systems are already accurate enough for two-dimensional joint angles, but estimating three-dimensional movement from a single camera view remains a challenge. In practice, lighting, camera position, loose clothing and the angle of the movement all affect accuracy.
Evidence at a Glance
| Technology | Best-studied use | Evidence so far | Main caveat |
|---|---|---|---|
| Robot-assisted gait training | Walking after stroke | Moderate certainty of benefit when added to physiotherapy | Ideal dose and long-term effect unclear |
| Robot arm training | Arm weakness after stroke | Large trial found no benefit over usual care | Possible benefit in specific subgroups not yet proven |
| Powered exoskeletons | Spinal cord injury | Safe supervised walking in small studies | Limited everyday use; few controlled trials |
| Digital programs with motion tracking | Chronic back and joint pain | Similar results to in-person therapy in a company-run trial | Needs independent trials; not suited to all injuries |
| AI movement analysis by camera | Exercise feedback and range of motion | Promising for 2D measurements | 3D accuracy from one camera still limited |
Questions to Ask Before Using a Rehabilitation Technology
- Has it been tested in people with my condition and stage of recovery, not only in healthy volunteers?
- Is it used in addition to my therapy, and who reviews my progress?
- Who will I contact if an exercise causes pain or a new symptom?
- Does the app or device have the regulatory status required where I live (for example, FDA clearance in the US or CE marking in the EU)?
- What data does it collect, where is it stored, and who can see it?
- Is it covered by my health service or insurer, and what happens if I stop using it?
Related reading on Helathora
Limitations
Rehabilitation research is difficult to run well. Patients cannot be blinded to whether they are using a robot, groups often differ in how much therapy time they actually receive, and outcomes depend heavily on therapist skill. Many device studies are small, short or funded by manufacturers. Evidence from stroke does not automatically apply to sports injuries, fractures or post-surgical recovery, and access to these technologies varies widely between countries and health systems. Finally, many products marketed as AI-powered do not publish how their algorithms were trained or tested.
The Bottom Line
Frequently Asked Questions
Can an app replace my physiotherapist?
For some conditions, such as chronic low back pain, remote programs supervised by a physiotherapist have produced results similar to in-person care in trials. That is not the same as an unsupervised app. Recovery after surgery, fractures or neurological injury usually needs hands-on assessment and a professional who can adapt the plan.
Is robot-assisted therapy better than regular therapy?
It depends on what is being treated. For walking after stroke, adding robot-assisted gait training to physiotherapy probably helps more people walk independently. For arm weakness after stroke, the largest trial found no added benefit over usual care.
Can people with spinal cord injury walk again with an exoskeleton?
Some people with specific injury levels can stand and walk with a powered exoskeleton after training, usually with crutches and a trained companion. These devices support exercise and standing but do not restore natural walking, and most users still rely on a wheelchair for daily mobility.
How accurate is AI exercise tracking on a phone?
Phone-based tracking can measure some joint angles reasonably well, especially in two dimensions and in good conditions. Accuracy drops for complex three-dimensional movements, poor lighting or awkward camera angles, so it works best as a guide that a therapist checks.
Official Sources
- World Health Organization. Rehabilitation: a key strategy in addressing noncommunicable diseases (policy brief), February 2025.
- World Health Organization. Rehabilitation fact sheet.
- Mehrholz J, et al. Electromechanical-assisted training for walking after stroke. Cochrane Database of Systematic Reviews, 2020.
- PEDro. Summary of the Cochrane gait training review.
- Rodgers H, et al. Robot assisted training for the upper limb after stroke (RATULS). The Lancet, 2019.
- Miller LE, et al. Clinical effectiveness and safety of powered exoskeleton-assisted walking in patients with spinal cord injury. Medical Devices: Evidence and Research, 2016.
- van Dijsseldonk RB, et al. Exoskeleton home and community use in people with complete spinal cord injury. Scientific Reports, 2020.
- Cui D, et al. Randomized-controlled trial assessing a digital care program versus conventional physiotherapy for chronic low back pain. npj Digital Medicine, 2023.
- The potential of computer vision-based marker-less human motion analysis for rehabilitation (review).
Medical disclaimer: This article is for general information only and is not a substitute for professional medical advice, diagnosis or treatment. Rehabilitation needs differ greatly between injuries and individuals. Do not start, stop or change a rehabilitation program or use a rehabilitation device without guidance from your doctor or physiotherapist. Helathora is not affiliated with any company or product mentioned.