Whereami: Uses WiFi signals and machine learning to predict where you are | Hacker News Reader

Whereami: Uses WiFi signals and machine learning to predict where you are

github.com

52 pointscrummy11 comments

It works surprisingly well. Very reliable and precise. I didn't expect it to be able to detect if I'm on my bed or on my desk, which are separated by barely 5 feet.

Why would I need “machine learning” for this?

These days machine learning,blockchain, artificial intelligence, privacy, smart things just seems to be buzz word for trivial things. And in academia if students generally mention these words so that they can get funding. Also if you are going to join company you can say "I did deep learning for this or that to impress the interviewer"

Obviously the core goal of the product is useless. But it is fun.

How would you solve this without fitting a statistical model?

Why wouldn't you use machine learning? This is simply using a random forest, nothing exotic or some overly complicated neural network.

A prediction is about a future event. The word you're looking for is "guess".

Basically if you can imagine some horribly invasive and unethical technology it means that someone will make it and then expect kudos from the community.

If you look at the github readme, it becomes clear that this technique can't infer _real world_ locations from wireless signals (which is already very much possible and in widespread use), it's applying abstract labels to different regions of signal strengths. Someone has to train it themselves with those labels.

It just knows, "I am fairly confident you are in 'bathroom' given the current patterns of signal strengths from the access points I can currently see".

This might be useful as a low-power alternative to proper localization, for the cases that it's useful to know what room something's in, but not necessarily an actual position.

So, if you wanted to automate things in your house to happen when you walk into a room with your smartphone, this would be less invasive than actually trying to triangulate a position or doing visual SLAM :-)

This tech is widely known and already widely used.

The A in A-GPS is short for Assisted. It means that in the mix are added known WiFi signals.

GPS (Signals from satellites) is already supported by WiFi signals that are stored in a database that is updated in your phones memory based on roughly your location. [1]

Why do you think a GPS location fix is so quick nowadays? Those satellites didn’t get any better…

So why do we even need gps since WiFi signals can get us really far already? There isn’t much WiFi on highways or in nature. Also, a Tomahawk missile should hit a target even if that target doesn’t have WiFi. GPS was invented for the military.

[1] https://en.wikipedia.org/wiki/Assisted_GNSS

> Why do you think a GPS location fix is so quick nowadays? Those satellites didn’t get any better…

The article you linked to doesn't seem to mention using WiFi location data for assisted GPS. It does point out that devices with an internet connection know their time and usually have a pre-seeded copy of the Almanac, both of which make a massive difference in speed of initial lock.

Assisted GPS refers solely to receiving the info the device needs to locate the satellites (i.e. the almanac) via another source, rather than directly from the satellites.

Receiving it from the satellites directly takes ages because the bandwidth of the communication channel is very low.

It has nothing to do with WiFi - the transport of the data can be anything (e.g. mobile network).

You are right that devices also use available wifi signals for location, but it's not AGPS.

What seems novel about this is the ability to learn more precise room-level locations.

Reply on news.ycombinator.com