Andyvision wandering around in a shop

Robots are now a constant presence in modern society but we usually don’t see them since they operate in backstage. Factories are more and more “inhabited” by robots manufacturing cars, computers and shoes to name but a few. We have robots in our homes, but they are disguised and we don’t perceive them, like some vacuum cleaners or cooking devices. Hence, when we recognize a robot wandering around we are still surprised. That is not the case in South Korea were people are getting used to see robots in the subway stations, post offices, hospitals and also in department stores.

At CMU researchers have developed a robot that can move around in a shop to look at shelves, track merchandize and send alerts if needed, like the shortage of a certain ware.

To do this researchers have embedded many technologies in the robot, including 3D vision, image recognition automatic learning. They are not new but their use in a coordinated way is new. Don’t underestimate the difficulties: can you tell by looking at a shelf what sizes of a certain shirts are there? The robots try to look at bar codes (but quite often they are hidden inside the shirt…) measure the length of the shirt and figures out how it has been folded to derive the size. It also checks the inventory data base to get hints on what it sees and to compare its calculation with what should be available in the store.
All these technologies used to be pretty expensive but their price went down considerably (and it is still going down) and it may reach a point when it might be easier to use vision to understand an ambient than place RFID tags on any object.

The robot, called Andyvision, wanders the shop and creates a “mental map” of the shelves and the various products in its first “rounds”. Once it has understood the lay of the land it starts its rounds to check if there is any shortage of a product on a shelf, if a product has been misplaced and so on. In practice it provides a real time inventory of the shop. So far it has been “at work” in the CMU store since mid May and by September the researchers hope to be able to draw some conclusions on its effectiveness and to evaluate the potential money saving made possible by a real time environment.

From this to helping customers finding a product it is a small step. And that can led to increase a shop revenues.

Interestingly, studies are showing that the trend for mixing robots and humans is gaining steam and some analysts consider this to be the reason why in the USA global productivity has increased since 2007 in spite of a decrease of total jobs. For the first time the shift from one kind of society (agriculture to industrial, industrial to services) has not resulted to a shift of jobs that overall rebalances the job destruction with job creation.

As manager of this blog I am getting more and more concerned by spam, computer programs posting comments to advertise products based on some words captured in the post. I am getting more and more of them and this tells me that the CAPTCHA code being used is getting more and more easy to solve by computer.

We are getting some 20 spams a day, from about 6-10 different sources, thus indicating that there are several computers that can decode CAPTCHA.

This is an opportunity for new companies to develop smarter CAPTCHA, like NuCaptcha.

The company has invented a CAPTCHA consisting of moving letters and is telling you which ones have to be use as a code, in the case of the picture on the left you are supposed to type the red letters (KSR) as they move along the screen.

However, researchers from Stanford have just announced that borrowing from the understanding of human vision they have developed a program that is able to capture the floating letters with an accuracy of over 90%, thus invalidating the NuCaptcha solution.

I have to say that more than once I have had problem in understanding the scrambled letters myself and knowing that a computer can do better than me in spotting the right combination makes me feel a loser.

Are we really getting close to Kurzweil singularity, estimated at 2025, when computers will be able to best humans in most areas and look more intelligent than us?