"AI-powered" is a buzzword used to describe everything from a toothbrush to a refrigerator to a frying pan. So, when a walking device claims to use AI for gait adaptation, it's understandable to be skeptical.
The VIGX Pi Plus has onboard sensors that measure your movements and onboard processing that adjusts the motor output to those measurements in real time, without a cloud connection. Whether that makes it "AI" or just "a really good control system" is a question of semantics. What is important is what it accomplishes.
The Problem With Fixed Assistance
To understand why dynamic assistance is important, let's first consider what it replaces.
A fixed-assist device, such as a knee brace with a spring or a passive ankle support, applies the same force regardless of your actions. However, real movement involves varying speeds, inclines, declines, uneven terrain, weight shifts, pauses for photos, and changes in gait due to surface transitions.
Fixed-assist devices struggle with these transitions, feeling mechanical because they are. Consumer-grade walking assistance devices have failed to address this issue, as they often feel like braces rather than extensions of your natural movement. Users discard them after a few weeks because the constant low-grade resistance is more bothersome than the assistance provides.
The VIGX addresses this by continuously adapting its assistance to your actions. Without this adaptation, the device would simply be a heavy brace with a motor.
The VIGX measures stride length, cadence (steps per minute), speed, and directional movement patterns. These data points are used to build a model of your individual gait.
What "Real-Time" Actually Means
A note on what "real-time" means in this context, as the term is often vague.
The device's control loop operates on the order of milliseconds. Sensors continuously take readings, the processor evaluates these readings and decides on a motor output, which the motor then adjusts. All of this happens rapidly enough that there's no perceptible lag between your motion and the assist.
Why this matters: a slower control loop would make the device feel jerky or delayed. You'd plant your foot, and the assist would arrive a fraction of a second later, after the moment when you needed it. The device would feel like it was assisting you with the previous step instead of the current one. You'd notice the lag, and noticing it would make the device feel mechanical rather than fluid.
A fast loop is what makes the assist feel like a natural extension of your movement instead of a delayed response to it. When you start a step, the assist is there at the beginning of the step. When you transition into a faster pace, the assist increases immediately, not after you've taken a few accelerating strides. This is a design detail that doesn't show up on a spec sheet but does show up in how the device feels.
What Adaptation Looks Like in Practice
When you first put the device on, it operates from a general baseline model. Within a few minutes of walking, it has enough data to begin applying a personalized profile. By the end of a first session, it becomes noticeably more responsive. Over multiple sessions, it continues to refine.
In week one, the device feels like a piece of equipment that's helping you walk. In week three, walking just feels easier. The mechanism that gets you from one to the other is the AI building an increasingly accurate model of your gait.
The "invisible" quality experienced users describe isn't comfort in the way a soft pair of socks is comfortable. It's the AI having modeled your gait well enough that the assist doesn't feel like interference. It just feels like easier walking.
A useful comparison: consider the first time you drove a manual transmission car versus the hundredth time. Initially, you're acutely aware of every clutch press, every gear shift, and every release. However, by the hundredth time, you're not consciously thinking about any of it. The car feels like an extension of your intentions. The mechanical complexity remains unchanged, but your relationship with it has evolved.
The VIGX follows a similar trajectory, but compressed into a few weeks instead of months. The change isn't solely on your side; the device is also adapting to you. You're both meeting in the middle. By the time the relationship is established, the device has developed a unique model of your gait and movement patterns, and you have a sense of how to use the assist that's specifically tuned to the device's response. It's a two-way calibration process.
The Activity Detection Feature
The AI doesn't only adapt to your gait; it also detects the specific activity you're engaged in.
Walking, running, and cycling each produce distinct sensor signatures. The device recognizes which activity you're performing and adjusts its assist profile accordingly. You don't need to set separate "running mode" or "cycling mode" settings. Simply start running, and the device adapts. If you stop running and start walking, it adjusts again.
This may seem like a minor detail, but it's one of the design choices that makes the device practical in real-world use. Imagine the alternative: a device that requires you to manually switch modes every time your activity changes. If you're walking around the block and the light turns green and you jog across, the device is still in walking mode, so the assist is ineffective. After you finish jogging and start walking again, you'd have to remember to switch the mode back. After a week of this, the device would likely end up in a drawer.
Transitions are where most assistive devices falter. The VIGX addresses this by eliminating the need for mode management.
A surprising example came from a friend who tested the device on a mountain bike. Although not primarily designed for cycling, the AI recognized the cycling motion and provided useful assistance. He successfully completed most of the way up a hill he'd always avoided. The point isn't that this is now a cycling device; it's that the AI is general enough to handle motion patterns it wasn't specifically designed for. If a motion pattern is similar enough to one the system can model, it will work with it.
Another example from a few weeks of use involved walking up a flight of stairs in the subway. Stairs differ from walking in cadence, leg lift, and push-off timing. Despite these differences, the device handles them without any input from me. I don't even think about it. The assist is precisely what I need, at the right time.
What the AI Does Not Do
Here's what the AI does not do:
-
It doesn't know your goals, such as training for an event or wanting a harder workout.
-
It doesn't know if you've just come back from a long break and want to take it easy.
-
It doesn't read your mind.
How This Compares to Other "Smart" Wearables
Most consumer wearables that claim to be smart are devices that record data and allow you to review it later. A smartwatch tracks your heart rate and displays a graph at the end of the day. The "intelligence" lies in the analysis, not the device's behavior.
The VIGX is different. Its intelligence lives in the device's real-time behavior. The VIGX's sensors feed into a control loop that adjusts motor output as you move. The output of the AI is not a graph at the end of the day but rather the difference between assist that complements your gait and assist that opposes it.
The primary purpose of all this technology is to eliminate the need for conscious thought. The ideal version of AI in a wearable is the one that goes unnoticed. You simply put on the device, set a level, and walk, hike, run, or ride. The device handles the rest. Sensors measure your motion, a processor constructs a model of your gait, and a motor adjusts accordingly. None of this requires your attention, and the design objective was to ensure that none of it ever does.
After six weeks of using the VIGX, I use it something like this now: I wear it on my walks, adjust the settings depending on how strenuous, or how big the hills are, and the AI does the rest, which is how it should be.
For more information about the VIGX Pi series, visit vigx.ai.
