Plurilock Next Gen Cybersecurity
Trending >

This ghostly singing robot from Japan will creep you out

singing robot

singing robot If it’s that time of day again when you’re looking to get creeped out a little, official sources say Alter, a singing robot, is your best bet.

On display at the Japan National Science Museum, Alter is the latest in robotic programming, displaying a human-modeled ghostly white face, arms and hands and a rods-and- wires frame involving 42 pneumatic actuators and multiple sensors for detecting object proximity, noise, temperature and humidity.

Most importantly, Alter is powered by an artificial neural network called a central pattern generator or CPG that takes in sensory readings and creates movement and sound responses, moving in a somewhat jerky but continuous motion and sending out sound patterns (okay, the thing is singing, get over it) in tandem with arm and finger movements.

Alter is the creation of researchers from the University of Tokyo and Osaka University, who consider Alter’s CPG a breakthrough in neural networking, one which relies on a simple process called spiking and burst behaviour to prompt movement and sound. “Until now, making androids talk or interact for 10 minutes was an incredible amount of hard work, simply to program something to react for that long.” says Osaka University’s Kouhei Ogawa. “Alter, moving for itself, can do so easily.”

The rhythm created by this build-up and release of movement and sound is eerie to say the least, but Ogawa is not concerned. “This time, Alter doesn’t look like a human. It doesn’t really move like a human. However, it certainly has a presence.”

Artificial neural networks mimic biological structures like the brain and central nervous system in order to not merely replay preprogrammed commands but to be able to analyze new information and essentially learn from it, creating more sophisticated responses. Your smartphone’s voice command recognition is a prime example of this. The field has expanded greatly even over the past half decade, as companies like Google and Microsoft have begun to see the potential in what’s referred to as “deep learning”.

In January, 2014, for example, DeepMind, a British artificial intelligence company with expertise in neural networking, was bought for $400 million (USD) by Google, who had also just hired University of Toronto computer scientist Geoffrey Hinton, nicknamed the ‘godfather of deep learning’.

The uses for deep learning span almost every field of inquiry. Take renewable energy, for instance. Biomass production involves a number of non-fossil- based organic matters like plants, timber and solid wastes that go through either combustion, gasification or decomposition in order to produce energy. The problem with biomass energy is that both the efficiency of the energy production and the emissions released depend upon the particular composition of the materials making up the biomass and the temperatures achieved in the process, both of which need to be carefully calibrated in order to devliver peak efficiency.

Enter neural networking, says a team of scientists whose research was recently published in the journal Fuel, whereby the complex and continually changing variables within the biomass can be quickly analyzed for optimum performance. “These neural networks must be continually fed,” says project lead and professor of Chemical Engineering at the University of the Basque Country. “As the results improve when broader case studies are incorporated.”

Below: Alter: a new type of robot [RAW VIDEO]

More Cantech Science

  •  
  •  
  •  

About The Author /

Jayson is a writer, researcher and educator with a PhD in political philosophy from the University of Ottawa. His interests range from bioethics and innovations in the health sciences to governance, social justice and the history of ideas.

Comment

Leave a Reply

Your email address will not be published. Required fields are marked *

RELATED POSTS

Cantech Alerts.

Timely picks from Canada's best analysts. 

F                                                                      
close-link