Gigapan comes to Google

Charmed Labs has announced a new collaboration with Google to integrate their gigapixel panoramic images with Google maps in order to create a new educational medium for cross-culture learning. I have presented the Gigapan technology in a previous post, Gigapan: High-resolution panoramas using consumer-level digital cameras. The system is now available for beta testing for early adopters willing to spend $279 to purchase the device; you probably have to supply your own digital camera but who doesn't own one these days. The registration form is here.

The following video introduces Gigapan and the educational project behind the new gigapan.org website where anyone can upload and share their own panoramas (in fact, there are already a large number of great panoramas available on the website and it is definitely worth a visit.)

Looj and ConnectR: iRobot's new robots

The company that brought us Roomba, Scooba, Verro, Dirt Dog and a number of military robots is at it again. Earlier today, iRobot announced two new robots: Looj and ConnectR.

Looj is a new robot on the family of cleaning machines that have been very successful for iRobot. Looj is a gutter cleaning robot that can clean a 60-ft section in 10 minutes and break up clogs. The Looj comes in 3 models, the 120, 130, and 150. Differences are in the accessories available with each model other than the price which starts at $99.99 for the 120 and ends at 169.99 for the 150.

The ConnectR Virtual Visiting robot is designed for telepresence applications. It comes equipped with a camera and microphones making remote interaction possible. The robot is not available for sale yet but a limited number of units will be available in what seems to be a beta testing program (or pilot program as iRobot calls it.) Communication will be performed via iRobot's Virtual Presence Network to ensure the security and privacy of users. ConnectR will retail for $499 but those who sign up for the pilot program will be given the chance to purchase the robot at the reduced price of $199.

If I remember correctly, a few years ago, iRobot was developing another robot similar to ConnectR but much taller. I can't recall the robot's name but obviously it was a product that never made it to market. In other words, this is the second time the company is trying to monetize the remote presence market (if such a market actually exists.) I'm curious to see if they will be successful this time around.



Images are copyright iRobot.

Optimal Group acquires WowWee

News today are that robot toy maker WowWee has been acquired by Montreal's Optimal Group for $65 million. The Hong Kong-based company has found much success in recent years with their extensive and popular line of toy robots including a variety of humanoids, a robot dinosaur, a robot snake, a robot panda and a robot dragonfly among many others. The new owner is determined to continue expanding in the robot business with new products and better distribution.

Read the press release here.

Improving stereo vision with monocular cues

Stereo vision is an essential component for any mobile robot. The most popular stereo vision systems available today estimate the depth of points by finding matches between two (or more) spatially separated cameras and performing triangulation given known camera parameters. This correlation-based stereo algorithm is easily implemented in real-time on traditional CPUs or specialized FPGAs.

Recent work is trying to address problems in stereopsis that result most often from a lack of scene texture; anyone who has used stereo vision for robot navigation is probably very familiar with this problem. Looking for a solution to this problem, researchers from Stanford University utilize machine learning techniques to learn the depth of objects using monocular cues and then use it to improve traditional stereo. CMU's Derek Hoiem has performed extensive research on the problem of recovering 3D structure from single images; you may already be familiar with his automatic photo pop-up application. In a series of conference and journal papers, Ashutosh Saxena and his colleagues have now shown how monocular cues such as texture variations and gradients, defocus, and color combined with correlation-based algorithms can be used to improve depth estimation.

The researchers are using supervised learning techniques to learn a discriminate, hierarchical Markov Random Field (MRF) model for combining monocular and stereo cues. They report significant improvements in depth accuracy over using only monocular or correlation-based stereo.

Saxena has posted the papers along with a number of examples demonstrating the approach here.

Image is copyright Ashutosh Saxena and Stanford University.

Bad Science

I always enjoy reading Ben Goldacre's weekly column in Guardian Unlimited about how science is misused (I'm trying to avoid using a stronger word here) in the mass media. Sometimes this is the result of a journalist's gross misunderstanding of the science due to his/her lack of knowledge in the subject. Other times, this is done on purpose in order to promote the interests of third parties whatever those are. The description on Guardian's website describes the column as,

Each week, Ben Goldacre skewers the enemies of reason. If you're a journalist that misrepresents science for the sake of a headline, a politician more interested in spin than evidence, or an advertiser that loves pictures of molecules in little white coats, then beware: your days are numbered.

If you have not read his column before, it is worth spending some time reading some (if not all) of his articles. Dr. Goldacre also writes the Bad Science blog which he updates much more frequently. He is also preparing a new book titled Bad Science (to be available early 2008).

New theme/template for our blog

Those of you who come to the blog to read the most recent articles (as opposed to the email and RSS subscribers) have probably noticed that since last night, we have updated the blog's theme/template. The new theme is Minyx based on a Wordpress theme by Mauricio Caballero; the blogger version was made available on B-Themes.

I am still tweaking the template to better suit our Artificial Intelligence and Robotics blog. So, expect things to be a little fragile over the next few days.

Robots for small businesses

CNN Money has posted an interesting article about how used robots are becoming affordable for small businesses. Industrial robots are prevalent in many factories most notably auto plants. Such robots are expensive, robust, require little maintenance and of course can repeat the same motion over and over again with extreme precision. The benefits of owning such robots vastly outweigh their cost but only for large manufacturers. Second-hand robots cost less and are now becoming well suited for the use by small businesses which can benefit from automation. The article cites the case of Blue Chip manufacturing that purchased two used robots for a fraction of the cost of new ones; the company was able to payoff the two machines in less than a year while saving over $60,000 in manufacturing costs.

Read the full CNN Money article, "Need more workers? Try a robot."

BDAGL: Bayesian DAG learning software release

Daniel Eaton and Kevin Murphy from the University of British Columbia have released a new Matlab/C/Java software package for Bayesian inference about fully observed Directed Acyclic Graph (DAG) structures using dynamic programming and MCMC.

I am pleased to announce the release of "BDAGL" (pronounced "be-daggle"), a Matlab/C package for learning Bayes net structures from fully observed data (discrete or continuous, static or time series). Its main novelty is that implements various algorithms for exact Bayesian inference of posterior features/ modes using dynamic programming. It also supports MCMC (on DAGs and orders) with various proposal distributions. Details are given at the URL below. Feedback is welcome.

You can download the software along with installation instructions, usage examples and references here.

CMU and the Google Lunar X PRIZE

There is a story posted on tgdaily.com saying that CMU will be going after the Google Lunar X PRIZE with a team led by Professor William Whittaker. Whittaker is the same person who managed CMU's DARPA Grand Challenge and Urban Challenge teams. He has lots of experience developing robots of all sizes and according to the article, he recently developed a new power-efficient rover for NASA.

I wonder if other Grand Challenge participants will follow suit and announce their entry in the Google Lunar X PRIZE.

[ via Slashdot.org ]

DARPA Grand Challenge book

Springer has published a new book in the Springer Tracts in Advanced Robotics series regarding the DARPA Grand Challenge. The book is a collection of papers written by the participating teams explaining the philosophy and technology behind their autonomous vehicles.

This book will serve as an authoritative, archival source for the DARPA Grand Challenge and is a must have for robotics students and researchers, since it describes the state of the art in perception, planning and control.

More information about the book, The 2005 Grand Challenge: The Great Robot Race, and how to purchase here.

The ICRA Robot Challenge

There is a new Robot Challenge currently being organized in conjunction with the International Conference on Robotics and Automation (ICRA). The new Robot Challenge will have space robotics as the main theme and the first competition will take place next year during the ICRA 2008 conference to be held in Pasadena, California.

For the time being, two events are scheduled: The sandbox, and the planetary robotic contingency. Details have not been finalized for the two events as the organizers are asking researchers who want to compete to help them define all the parameters. However, broad descriptions of the two events have been posted on the Robot Challenge website.

The sandbox will simulate a planetary rover mission in a small, 6x6 meter area covered in sand and rocks. Contestants will be asked to develop robots to perform autonomous navigation, mapping, data collection and object retrieval.

The robotic planetary contingency challenge is more ambitious. Organizers hope to simulate an unexpected problem during a space mission and ask participants to develop autonomous robotic systems that can quickly devise a solution using only existing resources. In this case, the sandbox will function as the planetary environment while a second area called the habitat will be used by the robotics team to develop their solution to the given problem. Some example missions include antenna recovery, base station repair and buggy diagnosis and repair.

Teams that might be interested in the Google Lunar X PRIZE might find the ICRA Robot Challenge a good place to test new ideas and technologies.

The strangeness of science

Interesting lecture by evolutionary biologist and Oxford Professor Richard Dawkins discussing the strangeness of science. This lecture he gave in 2005 as one of the speakers for the TED conference. But don't think that it is out of date in any way.

British Computer Society Machine Intelligence Competition

The Artificial Intelligence Group of the British Computer Society is holding the sixth annual machine intelligence competition open to all for no registration fee and with a cash prize and trophy.

The demonstration can be of either software (e.g. a question-answering system or a speech recognition system) or hardware (e.g. a mobile robot).

This competition will put on show real systems demonstrated live. It is hoped that the competition and the competitors, over several years, will provide a new interest and visible improvements in the development of machine intelligence.

The best demo will be selected by the AI-2007 conference attendees based on a 10-15 minute live demonstration. You can find more information about the competition including a list of past winners at the official website here.

[ Thanks to Richard Ellis for the tip. ]

Sir Arthur C. Clarke on Moon 2.0

A very powerful message by sir Arthur C. Clarke endorsing the recently announced Google Lunar X PRIZE. Video below,

Sebastian Thrun on Talking Robots

For their one year anniversary, Talking Robots interviewed the most successful robotics scientist in recent times, Stanford's Sebastian Thrun. If you don't know, Thrun is the man behind Stanley, the winner of the DARPA Grand Challenge; he is also the leader of one of the most ambitious teams in the new DARPA Urban Challenge. Thrun, of course, is a very prolific scientist having co-authored a large number of papers and books in addition to the numerous robots that he and his students have developed over the years.

During the interview, Thrun answers questions about the Grand and Urban Challenges as well as the present and future of mobile robotics. Notably, Thrun has written the book on probabilistic robotics describing the modern methods used for modeling and implementing real-life robots. Probabilistic robotics is an area that he helped pioneer with his very successful solutions to the problems of navigation, mapping and localization in the late 90s. Interestingly, he says that the future of robotics is mostly in software. He believes that hardware has matured to the point that other than lowering its cost, there is little to be done. Software, on the other hand, still has a long ways to go especially in real-time planning under uncertainty, semantics, and of course machine learning. I actually agree with him.

You can download the Podcast here.

Also, the Talking Robots people are asking their listeners to take a small survey in order to help them improve the already very well done and successful Podcast. More information here.

Google Lunar X Prize

Google and the X Prize Foundation announced today a new prize for the first privately funded team to land a robotic rover on the Moon; they call this Moon 2.0 as opposed to NASA's original Apollo Moon missions. Google will pay $30,000,000 (that's thirty million dollars) to the winners; the money will be distributed as follows: $20 million for the grand prize, $5 million for second prize and another $5 million in bonus prizes. The bonuses are for those teams that not only land a rover but perform well in tasks such as discovering water ice on the Lunar surface. Teams are given the hard deadline of December 31st 2014 for claiming the X Prize. The press release explains the goals of the Google Lunar X Prize,

Private companies from around the world will compete to land a privately funded robotic rover on the Moon that is capable of completing several mission objectives, including roaming the lunar surface for at least 500 meters and sending video, images and data back to the Earth.

But talk is cheap, so check out the promotional video from the X Prize foundation explaining the Google Lunar X Prize,


This sounds like a very exciting project and I am kind off jealous of those people who will have the privilege of participating. I will definitely be following this very closely.

Google Street Views illegal in Canada?

There is an interesting article on CBC's website discussing Canada's privacy commissioner Jennifer Stoddart's look into the privacy issues surrounding Google's high resolution image mapping service. Stoddart claims that it is possible that Google is violating the country's privacy laws by capturing high resolution images of unsuspecting citizens.

We had previously talked about Google's street-level mapping service and mentioned the thorny privacy issues especially considering websites such as streetviewr.com. Interestingly, the image data used for Street Views is collected and compiled by a Canadian company, the Calgary-based Immersive Media.

For the time being, Google Street Views is only available for US cities but I am certain they plan to expand abroad sooner than later.

In a letter directed to Google and Immersive Media, Stoddart said,

I am concerned that, if the Street View application were deployed in Canada, it might not comply with our federal privacy legislation. In particular, it does not appear to meet the basic requirements of knowledge, consent, and limited collection and use as set out in the legislation.

I don't know what others think about having photos of themselves posted online without their consent, but I know I don't want my photograph there, regardless of the activity I may be doing at the time the Immersive Media car drives by. Google can use face detection technology to detect and blur the faces of all the people appearing in the images. There are lots of algorithms that can do this job reliably and efficiently. Doing this would definitely comply with the company's “Do no evil” motto.

Robot pet Pleo delayed (again!)

In an email titled "Battery Solution Results in a Delay," UGOBE's Caleb Chung announced that Pleo will once again be delayed and it won't ship for a currently unknown amount of time (but will be availably for the holidays.) The delay is apparently caused by an upgrade to the robot's battery system which will now be easy to remove and come with an external charger. Specifically, Caleb Chung says,

During pre-production testing, the UGOBE team encountered some significant problems with Pleo's battery life. We solved these problems in a way that I think you will appreciate: all Pleos, including First Hatch Pleos, are now designed to have an easily removable battery and an external charger for the battery. That means that with multiple batteries and chargers, you can keep Pleo going for hours on end. However, implementing this design change will delay production.

The good news are that UGOBE will make every effort to have the robots delivered by the holidays; it is not clear if this will be the case only for those who have already pre-ordered one for $350 or the robot pet will actually be available at the local stores at the same time (I would guess not.) Luckily, Caleb will be making more videos of Pleo in an effort to continue their ridiculous marketing strategy.

I know how much you are looking forward to having Pleo in your life so I know this delay is frustrating. At the same time, I hope that you are as excited as I am about this design change. I'll post some video or photos soon, so you can see it yourself.

I wonder which product will be delivered first: Pleo or Duke Nukem Forever.

Wheelchair control using a Brain-Computer Interface

Audeo is very interesting technology from Ambient Corporation that allows a user to control a wheelchair with thought alone. The company also claims that their neurological interface is advanced enough to generate continuous speech directly from a patient's thoughts giving him/her the ability to speak once again. This all sounds very interesting but I think that people don't realize that this is not a device that works like "traditional" BCIs.

The Audeo doesn't actually read your thoughts in the sense that it reads and decodes brain waves. The device is placed around a person's neck and it monitors the electrical signals sent to the patient's vocal cords that would normally cause the generation of speech.

The Audeo is being developed to create a human-computer interface for communication without the need of physical motor control or speech production. Using signal processing, unpronounced speech representing the thought of the mind can be translated from intercepted neurological signals.

By interfacing near the source of vocal production, the Audeo has the potential to restore communication to people who are unable to speak. The proposed solution is a featherweight wireless device resting over the vocal cords capable of transmitting neurological information from the brain. Using data analysis, this information can be processed into synthesized speech or a menu selection capable of conveying the basic necessities of human life.

The following video shows the founders describing how the Audeo works,


Finally, here is the company's video demo of a wheelchair controlled by “thought” using Audeo technology.

Microsoft forms alliance with Tmsuk robot manufacturer

Microsoft and Tmsuk have announced a new partnership that will have Microsoft's Robotics Studio powering Tmsuk's robots. Tmsuk is a Japanese robot manufacturer that has been building robots since 1992. The large collection of robots Tmsuk has developed ranges from traditional wheeled robots to humanoids; the robots are designed for a number of different applications including health-care and office jobs, i.e., robot receptionists.

Tmsuk joins the large number of robot manufacturers who are providing drivers for the Microsoft Robotics Studio (MSRS), Bill Gates's initiative to create a standard platform for robot programming and development. Even though Microsoft only has a small number of engineers developing MSRS, they have achieved a large penetration in the market in a bit more than a year since their software development suit has been available to the public.

Gates is obviously taking this initiative very seriously. Microsoft has no problem using their influence in the industry to dominate a new market. The open source Player/Stage initiative is probably feeling the heat at the moment but they are also reporting more than 60,000 downloads. The battle has not been decided yet because even though software seems to progress quickly, hardware hasn't moved forward by much this year. Still, 2008 will be a critical year on deciding who will be providing software for the consumer robots that are expected to start dominating the market over then next decade.

Leg-Wheel hybrid walking robot

Last month, I wrote about JPL's ATHLETE robot which has six limbs that can be used as both legs and wheels. ATHLETE is a large robot designed to aid astronauts during planetary missions. Japanese researchers have also been working on similar hybrid walking robots but for less ambitious applications. Take for example the four-legged Roller-Walker robot from the Hirose-Fukushima robotics lab. This is a small robot that can operate in two different modes, walking and roller-skating. The project web page describes it as,

Proposed hybrid mobile robot named "Roller-Walker" is a vehicle with a special foot mechanism which changes between feet soles for the walking mode and passive wheels for the wheel. Roller-Walker can utilize the installed actuators for walking, so additional weight is very light. The wheeled locomotion is based on the same principle of roller-skating.

The following video will give you a better idea as to how the Roller-Walker works (looks very cool.)



[via Robots.net]

Brain-Computer Interfaces dangerous?

There is an article posted on Wired.com that makes the case that Brain-Computer Interfaces used for game-playing over long periods of time may actually be harmful to people. Recent advances in technology are making it possible for companies such as Emotiv Systems and NeuroSky to release BCI kits to the public while game development companies are showing an interested in developing BCI-based games centered around these gizmos; I bet the recent success of the Wii and its strange new controller is giving companies the idea that consumers are interested in new game controllers so everyone is looking for the next big thing. So far, BCI-based applications have been used mainly in the medical sector for rehabilitation and patient treatment.

Scientists suggest that there might be some serious side effects associated with the use of BCI-based game controllers because of their long-term use something not common in medical applications. According to the article,

For example, the devices sometimes force users to slow down their brain waves. Afterward, users have reported trouble focusing their attention.

"Imagine that somebody uses a game with slow brain-wave activity and then drives a car while still in that state," says Niels Birbaumer, a leading independent researcher in medical applications of BCIs. "You could have an accident. I think it's a rare possibility, but it should be tested before people do this."

I am no brain science expert and so I am not so sure what to make of the claims in the article; it is possible that it is meant to generate fear more than inform. Scientists interviewed in the article are polarized. There is no hard evidence at the moment that support or contradict the hypothesis that BCIs are harmful. I do agree with the article's suggestion that in the end, there has to be a valid scientific study to evaluate the potential of dangerous side effects for consumer-level BCIs.

Science lectures online

If you are looking for science lectures as given by some very prominent scientists, then you should head over to Videolectures.net that hosts a great collection of videos. There seems to be a huge bias towards computer science and more specifically machine learning but there are some good gems in the collection including lectures by David Mackey, Arnaud Doucet, and Andrew Blake among many others. The site's mission as stated on their Contact Us page is,

to offer everybody a free access to high-quality scientific video lectures thus expanding educational opportunities, promoting scientific achievements, and facilitating the creation, use, and re-use of Videolectures.Net materials.

Notable collections include the lectures from the 9th Machine Learning summer school and AAAI video competition; but keep exploring because I am certain you will find many interesting lectures to keep you entertained for long periods of time. BTW, I find the Tutorial videos a good starting point for familiarizing myself on new areas of research or just brushing up on some older material.

Context aware image resizing

Another excellent piece of work presented at SIGGRAPH 2007 is Shai Avidan and Ariel Shamir's context aware image resizing paper. The two scientists describe in their paper how images can be effectively resized removing those pixels containing the least amount of information. They define an energy function computed for each image pixel and then find connected regions of low energy pixels that run from top to bottom or left to right; they call these connected regions seams. Seams are those regions that can be removed without much loss in the image's content. The paper abstract better describes how this parameter-free method works utilizing a new image operator called seam carving.

Effective resizing of images should not only use geometric constraints, but consider the image content as well. We present a simple image operator called seam carving that supports content-aware image resizing for both reduction and expansion. A seam is an optimal 8-connected path of pixels on a single image from top to bottom, or left to right, where optimality is defined by an image energy function. By repeatedly carving out or inserting seams in one direction we can change the aspect ratio of an image. By applying these operators in both directions we can retarget the image to a new size. The selection and order of seams protect the content of the image, as defined by the energy function. Seam carving can also be used for image content enhancement and object removal. We support various visual saliency measures for defining the energy of an image, and can also include user input to guide the process. By storing the order of seams in an image we create multi-size images, that are able to continuously change in real time to fit a given size.

The following video showcases the technique,