The wine-tasting robot

The evolution of information technology that imitates the five human senses of sight, hearing, touch, smell, and taste is important in the advancement of a ubiquitous society. Undoubtedly one of the leaders in this area is NEC System Technologies which continues to push robotic development as can be seen from these stories (here, here and here) on its technologies, all of which seem to be able to be shoehorned into the tiny PaPeRo form factor. More recently, the research has moved from areas such as personality, social skills, companionship and English-Japanese language translation to technologies that support the prevention of lifestyle-related diseases through dietary therapy. Last year we saw Papero transformed into a “health food adviser robot” which analysed food using infrared spectrum analysis and did so WITHOUT opening the packet. Not only did it report on the fat and sugar content, the robot actually identifies several types of cheese (Edam, Gouda, Camembert, etc.), meat products (ham, bacon), and bread (pain de mie, baguette, croissant,). This year they set out to build a robot that could differentiate between different types of wine, a far more exacting task using the same technology. The new wine-tasting robot can now discriminate between wine types but the sensor is much larger, needs to touch the surface of the wine and must be cleaned for each tasting.

It works by transmitting infrared rays at the food and measuring the degree of absorption of certain wavelengths (absorbance spectrum). The absorbance spectrum gives out different wave shapes for different foods. When a certain molecule is present in a food, a ray of a specific wavelength is absorbed, so the robot can estimate the major components, such as sugar and fat, and the quantities of these components present in a food by consulting its food information database (containing data on all the foods it has previously “tasted”).

Adapting that technology to wine-tasting is difficult because differences in absorbance spectra among different types of wine are strikingly smaller than those among other types of foods. It is hence more challenging for the robot to discriminate between different types of wine than it is to differentiate between other types of food.

This was tackled on two main fronts – the development of a more acute infrared sensor with a wider spectrum that covers the range from near-infrared rays to the mid-infrared spectrum, and a new discrimination algorithm focusing on the extracted point of maximum difference between the spectra of the wines being tasted.

The NEC team continues to refine the process with the intention of creating a sommelier robot that can taste wine, and recommend wine suited to a customer's taste by carrying on a dialogue with him/her. Like a human sommelier, the robot asks a minimum number of questions in order to narrow down the choices among the best wines.