When It Rains, It Trains (AI)

Volume 29 October 13, 2025

From the Editor

It’s been raining so much in Phoenix this past month that I half expected my gait analysis sensors to start growing mushrooms. Meanwhile, AI researchers are still flooding the literature faster than our storm drains can handle it.

This week’s issue features robots that take coaching cues from therapists, wearables for balance rehab, and a new home stroke system that watches your form like a hawk. All great stuff.

But in the background, ongoing trade wars are starting to cast a shadow over AI hardwar, especially GPUs and cloud infrastructure. If pricing or supply chains shift, we could see downstream effects on everything from smart rehab devices to the servers running clinical AI tools. So while the tech keeps improving, affordability might become part of the conversation.

With the weather looking stormy both in Phoenix and the global trade market we hope everyone was able to stay safe!

📰 This Week's Highlights

AI-Powered Home Stroke Rehab System Uses Video + Wearables for Feedback

Preprint: September 27, 2025 (arXiv)
Researchers have developed a home-based stroke rehabilitation tool that uses a combination of RGB-D cameras, wearable sensors, and a new AI model (RAST-G@) to assess movement quality. Built on a spatio-temporal graph convolutional network with transformer-based attention, the system scores upper-limb exercises with feedback comparable to that of a physiotherapist. Their new NRC dataset—validated by licensed PTs—includes common ADLs and ROM tasks, providing the model with real-world relevance.

🔗 Read the preprint (6 min)

Why this matters for PTs:
Remote rehab often lacks clear, consistent feedback. A reliable AI that scores movement quality at home could help extend therapist oversight without increasing workload, particularly for patients with access challenges. This kind of tool could also aid triage, letting PTs focus on those who need the most in-person care.

Rehab Robots Learn from Human Therapists—Not Just Patients

Preprint: October 6, 2025 (arXiv)
A new system out of Germany aims to make assist-as-needed robotic rehab more responsive—by learning directly from therapists. Instead of programming rigid rules, the researchers captured subtle “corrective forces” from therapists and encoded them as waypoints in a neural network. The robot then adapted its path using a shape-deforming policy, balancing guidance with patient freedom. In telerobotic experiments, the result was smoother trajectories and less need for ongoing therapist correction.

🔗 Read the preprint (7 min)

Why this matters for PTs:
Robotic rehab can’t replace your clinical reasoning—but it might be able to learn from it. This approach keeps therapists in the loop while offloading repetition to the robot. It could help scale neuro or ortho rehab in settings with staffing limits, while still delivering individualized, evidence-informed motion support.

Smart Wearables Are Advancing Fall Prevention in Older Adults

Published: September 30, 2025 (JMIR Rehabilitation)
A newly published review in JMIR Rehabilitation evaluates smart wearable tech for balance rehabilitation in fall-prone older adults. The article covers a range of devices using inertial, pressure, and motion sensors, along with real-time biofeedback strategies like vibrations or sound. Several systems show promise in improving postural control, though gaps remain in standardizing metrics and proving long-term efficacy in clinical settings.

🔗 Read the article (9 min)

Why this matters for PTs:
Wearables are no longer just fitness trackers, they’re turning into balance tools. This review gives PTs a grounded look at what’s currently available, how it’s being used, and where the evidence stands. It’s helpful intel when deciding whether, and howto integrate wearables into your fall-prevention programs.What’s New in ManagePT

Whats new in ManagePT?

New Feature Preview: ManagePT Copilot Is Coming
ManagePT is gearing up to launch Copilot, an AI-powered assistant built right into your documentation workflow. Think of it as a smart sidekick that drafts SOAP notes, suggests goals, offers treatment ideas, and even screens for red flags based on your evaluation findings. It’s designed to cut down on admin time without dumbing down your clinical reasoning.

A limited alpha pilot is starting soon, and there’s now an official waitlist live on the ManagePT website. If you want to be part of the first group to try it, that’s the place to sign up.

Bonus Reads

• The Verge: "Can AI-Powered Exosuits Make Rehab More Personal?"
  A quick look at how lightweight, sensor-driven exosuits are being trained with patient-specific AI models and what that means for gait rehab.

• Nature Digital Medicine: "Predictive AI in Outpatient Care: Too Soon or Right on Time?"
  This commentary weighs the risks and rewards of using AI to forecast functional decline, including use cases for therapy triage and discharge planning.