How I searched, what I found, and why being wrong four times made the project better.

You can find my published literature review HERE and the associated bibliography HERE.

When I started this review, I thought I had a pretty simple story: nobody had used haptic guidance to help a blind runner navigate at running speed. That was going to be the gap my project filled.

Six weeks and fifty sources later, that story had changed a lot. What replaced it was more complicated, but also more useful. This post is about both parts of that process: how I actually searched, and what I learned from it.

How I searched

I want to explain the method honestly, because a literature review is only useful if someone can understand how you got your answers.

The platform. I used Google Scholar from July 15 through August 31. At first, my searches were broad. I mixed terms for the people I was studying (blind, visually impaired), the activity (running, track), the type of feedback (haptic, vibrotactile, tactile belt), and the goal (navigation, guidance, lane keeping). Later, the searches became much more specific: author names, exact paper titles, named systems, and even older phrases like “blind athlete guidance device,” because papers from ten years ago did not always use the words researchers use today.

Following the trails. Every useful paper led to more papers. I checked its reference list to see what research came before it, and I used Google Scholar’s “Cited by” feature to see what came later. A 2024 review of hands-free haptic navigation devices was especially helpful as a map of the field, but I still went back to the original papers whenever possible to make sure the review had described them correctly.

Sorting what I found. I kept track of who the participants were (blind, visually impaired, or blindfolded sighted people — which are not the same thing), how fast they moved, where the feedback was placed on the body, whether the cue reacted after an error or warned the person ahead of time, whether a human operator was involved, and what the researchers actually measured. Peer-reviewed papers count as evidence. Preprints are labeled as preprints. Commercial products and news stories are useful background, but I do not use them to prove that my project is new.

What I can’t claim. I did not keep a full log of my earliest searches, so I cannot honestly say exactly how many search results I screened. What I can say is that I kept fifty separate works and read several of the most important papers in full. The review was done by one person, only included English-language sources, and did not use a formal bias-rating tool. The search is also dated August 31, and this field is moving fast, so I need to keep going back and checking for new work.

Being wrong, repeatedly

This is the part of the process I am most glad I kept track of. The gap did not shrink just once. It got smaller every time I searched more carefully.

My first version said nobody had done this before. Then I read the closest papers more carefully and found that haptic systems had already been tested at running speed, although not in the exact way I wanted to test them.

While checking the bibliography, I found a 2022 waist-worn system that had been tested with a visually impaired runner that my original search had completely missed. A deeper search then found BLINDTRACK, a European project that built an eight-motor belt for independent track running in 2015 and tested it with six visually impaired volunteers. I also found Brazilian researchers who tested vibration-based course correction with visually impaired sprinters.

And one discovery really hurt: a 2017 study I had been calling “walking pace only” had actually put a blind athlete through three laps at real running speed.

I also found mistakes that were not about the main gap at all. One result I had credited to a paper actually came from its 2015 follow-up. A veering measurement I had written in degrees was really measured in meters. And one paper I said had not tested route-following had, in fact, tested exactly that.

I corrected each mistake and wrote down what changed.

None of that was fun to find. But finding it was the point of doing the review.

So what is the gap now?

The gap I can now defend is this: as of August 31, I did not find a peer-reviewed controlled experiment that compared an accelerating, early-warning turn cue with a simpler cue on a vibration belt worn on the torso, while visually impaired participants were running, and measured both how well they navigated and things like mental workload.

That sentence has a lot of conditions in it, but each one matters because somebody has already tested a version without one of them.

My project is not the first haptic running system, the first vibrating belt, or the first attempt at independent track running. What may be new is the exact experiment: comparing a specific signal against a control signal, with the people the system is actually meant for.

I also stopped trying to name one “closest paper.” There really is not one. BLINDTRACK is closest to the full system. The Brazilian and Italian studies are closest in terms of visually impaired runners. The 2022 skin-stretch system is closest in waist placement. The vision-and-glove systems are closest in portable sensing.

Another thing I learned is that a research gap does not have to be huge to matter. At one point, I thought that because the gap had become so specific, there might not be enough original work left for me to do. After talking with a few experienced researchers, I realized that entire PhD dissertations can be built around questions narrower than the one I am exploring.

What the field already knows

The existing research gives me useful starting points, but not rules I should blindly copy.

In one major torso-localization study, six- and eight-position layouts produced high identification accuracy and performed better than a denser twelve-position layout. The belly button and spine were also stronger reference points than the sides.

People can also tell which direction a vibration moved even when the spacing is too small for them to identify the exact motor. That is encouraging for a cue that is supposed to show drift rather than ask someone to name a specific motor.

Running makes vibration harder to notice, especially when several motors fire together or when a signal stays on continuously. Staggered patterns seem to work better. The studies also give me reasonable starting ranges for vibration frequency and pulse length, but those are values to test, not specifications to copy. A small coin motor under a shirt on a moving runner is not the same as a carefully controlled lab setup.

Both ideas that started the project still look worth testing, but the literature changed how I would test them.

The fish-inspired drift cue should probably stay mostly quiet: no vibration while the runner is centered, then a clear correction as the runner moves farther off course. If the cue is “graded,” only one feature should change — location, intensity, pulse rate, or the number of active motors — instead of changing several things at once.

The bat-inspired approach cue also survived, but with an important warning. An accelerating cue might look better simply because it gives the runner more warning or more vibration overall. So I need to compare it with a fixed-rate cue that starts at the same time and gives roughly the same amount of vibration. The changing rate should be the main difference.

Touch also is not automatically easier for the user. Several studies found that tactile guidance created less workload than audio, but another study found that adding vibration on top of audio actually made workload worse. My takeaway is simple: keep the number of signals small, and use touch for information that is not already being delivered another way.

What changes

The review changed four parts of the project.

First, the novelty claim is now specific and tied to a date.

Second, the bench study will include a short baseline test with the real haptic belt before participants exercise, and I will describe that study as an exploratory pilot rather than something more conclusive.

Third, the Stage 3 trials will include a third control condition and will record exactly when each cue was triggered, so I can separate a bad signal from a badly timed signal. The accelerating cue will be compared with a fixed-rate cue that starts at the same time and gives about the same total vibration, as well as with a single cue at or near the turn.

Fourth, I will stay precise about who took part in each study. Blindfolded sighted volunteers can help test the hardware and compare signals, but testing with visually impaired runners is ultimately necessary to know whether the system works for the people it is meant to help.

Final thought

The literature review did not kill the project. It made the project more specific.

I no longer need to ask whether haptics can guide a blind runner, because previous work has already shown that they can.

The better question is whether a carefully designed torso signal — especially one that warns about a turn before the runner reaches it — can improve turn timing and keep the runner on course without adding too much mental workload.

That is a question I can actually test, and it is the one I want the project to answer.

This post is part of an ongoing research series at runlikeafish.blog exploring biomimetic approaches to assistive technology for visually impaired runners. If you’re a visually impaired runner, a guide, or a researcher working in this space, I’d love to hear from you. You can reach me at brooke@runlikeafish.blog