Robot Builder Home

Robots: Learning

The latest episode of the Robots podcast interviews two experts in robot learning. Andrea Thomaz from Georgia Tech studies how humans teach and how humanoids learn with the goal to improve human-robot interaction. Among other things, Thomaz has designed her own humanoid robot Simon shown in the video above (have a look at this non-youtube video to see Simon's expressive ears in action). Our second guest Sethu Vijayakumar from the University of Edinburgh has done an astonishing amount of work spanning the fields of statistical machine learning, human motor control, Bayesian inference techniques and computational neuroscience. In the interview Vijayakumar speaks about the similarities and differences between human and machine learning, and explains how robots can learn about their environment through physical interactions. Read on and tune in!

Consciousness May Not Require Big Brain

In an appropriate follow up to yesterday's article on IBM's large scale cortical simulation, comes news from brain researchers at Queen Mary, University of London and Cambridge University. Lars Chittka and other researchers, who have been studying both computer simulations and animal brains, now believe that brain size is not as significant as previously believed with respect to general intelligence and consciousness. "No one would seriously suggest that a bigger computer is automatically a better computer", say the researchers.

We know that body size is the single best way to predict an animal's brain size. However, contrary to popular belief, we can't say that brain size predicts their capacity for intelligent behaviour. In bigger brains we often don't find more complexity, just an endless repetition of the same neural circuits over and over.

The larger amount of IO needed in larger animals for things such as more muscles and higher resolution sensors may be the main driving factor for needing larger brains. The researchers say it's possible significant intelligence and consciousness could be present even in insect brains. This fits in with other recent research showing that, despite their tiny brain size, bees could match the best machine-based face recognition algorithms. For more, see the full text the researchers report, Are Bigger Brains Better (PDF format), or listen to the BBC Radio interview of Professor Lars Chittka on BBC Radio 4. CC-licensed image of honeybee by flickr user Agent Opie

Real-Time Cat-Level Cortical Simulation Achieved

IBM announced today a new advance in the simulation of biological brains on computer hardware. As part of the DARPA SyNAPSE project, the Cognitive Computer Team at the IBM Almaden Research Center has achieved a large-scale simulation based on a new algorithm, called BlueMatter, that synthesizes neurological data.

BlueMatter, a new algorithm created in collaboration with Stanford University, exploits the Blue Gene supercomputing architecture in order to noninvasively measure and map the connections between all cortical and sub-cortical locations within the human brain using magnetic resonance diffusion weighted imaging.

The result achieved was real-time simulation of a cat-level cortex containing 1 billion spiking neurons and 10 trilling synapses. The simulation required 147,456 CPUs and 144 terabytes of memory. For more on this project, see IBM Research Center Cognitive Computing Manager Dharmendra S Modha's Cognitive Computing Blog.

Random Robot Roundup

The Shadow Robot Company let us know about a great funding opportunity in the EU called ECHORD. Get your robot project proposal in by 1 Dec for a chance at 12 months funding for a staff plus hardware. The PlasticPals blog has some photos and information on the Russian AR-600 humanoid robot. If you're in need of some Java free software for most commonly available servo controllers, check the newest version of ServoMaster. Joanne Pransky, the Robot Psychiatrist, wrote to tell us about a new "low-cost" industrial robot arm from ST Robotics, the R12-500 Firefly. If you buy one, maybe you can combine it with MobileRobots, Inc. new MT400 "build-a-bot" autonomous robot base. Amber Johnson writes, "We just posted an article: 10 Unbelievable Robots Transforming Medicine". Rog-a-matic sent a link to a giant beetle robot built in Japan. And the Swirling Brain spotted stories on tiny sea-going robots, an update on the Robopocalypse film by the author of How to Survive a Robot Uprising, the worlds first Arabic-speaking interactive robot, and EVOLTA, a running, jumping, toy robot. Know any other robot news, gossip, or amazing facts we should report? Send 'em our way please. And don't forget to follow us on twitter.

Roomba Pac-Man

The Research and Engineering Center for Unmanned Vehicles (RECUV) at the University of Colorado at Boulder has been developing software that helps robots form ad-hoc networks and distribute cooperative control of their operations. Some of the individuals at RECUV decided to create a cool demo on their own time to show off what their software can do. They've implemented a real-life version of Pac-Man using Roombas. They are quick to point out that despite the fact that the Blinky, Inky, Clyde, and Pinky Roombas seem determined to kill the Pac-Man Roomba, all the robots are actually quite safe. This is because, they say, all are "instilled with the Three Laws of Roombotics". And we'll note that the software is safe to download and use too because it protects your rights to examine, use, modify and distribute under the GNU GPL. There's one more video after the break with details on the waypoint navigation software used in the demo.

Last Refreshed 11/22/2009 8:02:34 AM