|
From: Colin Mayhew <colinmayhewphd@yahoo.co.uk> I can assure you that the Cooper project is a real and very tangible one. Your suspicion is perhaps understandable because the leaps we've made are rather significant compared to the current state of commercial AI. As Mr. Clarke wrote in Technology and the Future, "Any sufficiently advanced technology is indistinguishable from magic." What's important to remember in this famous quotation is not that the technology becomes magic, but rather that technology seems magical only to those who don't understand the details or are not knowledgeable of the history of a technology's development. It's for that reason that I've placed notes online and have included videos from different stages of the project. Have you seen videos of people interacting with the Kismet robot? That robot uses a fairly simple emotional model, yet people bond to it and treat it as a 'living' creature! It has become something magical from bits of aluminum and electrons whizzing inside silicon. Your experiences in the research sector I'm sure have shown you how disconnected the public can be from the realities of technology. There are autonomous machines (be they in medicine or oil well drilling) so removed from our daily lives that when we finally learn of them, we are shocked and amazed---far more so than had we followed the gradual steps and wrong turns the engineers made developing and finessing the technology. This project is real, and it, and the systems I've developed for it are going to change the way we live our lives. The most recent software revision I've tested on the robot has some powerful reasoning capabilities, a large step more powerful and versatile than that employed on the robot when I recorded the videos you may have seen online. They are perhaps powerful enough to seem like magic, but both devil and the angel of creativity are in the details. Soon enough, these little creatures will be animating the robots all around us and making our lives safer and more fulfilling. Regards, --- "Chris S." <chrisks@UDel.Edu> wrote: > Is your |
|
From: Colin Mayhew <colinmayhewphd@yahoo.co.uk> Hello Peter, Thanks for your enthusiasm. I'm not sure I completely understood your mail, but perhaps if I ramble on some, I'll respond to your comments by accident. The robot was extensively modeled prior to its realisation. The control systems were simulated and tuned in Simulink and a physics-based modeling package developed at mit (http://www.ai.mit.edu/projects/leglab/simulations/ The robot has an internal model of itself which it uses to update a running-guess as to it's center of gravity (CG). This guess is then fed back into the dynamics algorithm's world state as a part of the balancing feedback system, and helps determine foot placement. The shadows in the lighting test actually were very useful from a visual-computation point of view. Because the on-board lights were offset from the robot's cameras, a slight border shadow is visible to the cameras at the edges of objects, similar to the horrible shadows that outline people when you take photos with a flash. This border shadow actually makes object identification easier because the low-level edge detection algorithm reports such a strong response. The visual system is based on what we know of our own biological vision, namely that recognition is layered from many simple "behaviors," such as edge-detection, vertical and horizontal line detection, image brightness, region color, etc. A strong response to one of the lower-levels of layers, such as edge-detect have very strong emergent effects on the higher levels. I was pleasantly surprised by the robot's nighttime performance. I would be speaking falsely though if I said that the head-lamp's placement was planned from the outset to aide the visual system. There's another video on the site which may give you a better sense about the way this visual tracking works on the robot. Best Regards, Colin
--- peter@sgcib.com wrote: > |