Engadget editor shows his ignorance when it comes to robots

It is good that blogs are allowing the average person (like yours truly) to play journalist on his/her spare time, but it is becoming a nuisance when people try to write about things they don't understand. Take for example Engadget editor Nilay Patel who recently wrote the article "Robot lawnmower kills Danish man." The article reports on the unfortunate event that took place on Wednesday in Ballerup, Denmark, when a 45-year old municipal employee was killed in a tragic accident while operating a remote controlled lawnmower. Now, I understand that Engadget caters to an audience with the IQ of a turtle and so the writing style tends to be rather moronic as each post tries to be funny or sarcastic in some way. But Nilay may have crossed a line today when he tried to be funny writing about this tragedy. The line he crossed is that of showing no respect to the family of the victim including his 3 children who are now grieving for their loss. If I were you, I would consider apologizing.

In addition and unfortunately for you Nilay, this tragic accident is not the beginning of a robot revolution. The remote controlled Dvorak Spider lawnmower that the worker was using is just that: a remote-controlled machine. It is not a robot. It is no more a robot than your car. The Dvorak Spider may have some electronics to perform low level control so that the vehicle moves smoothly but it hardly does anything for itself; any modern car has such subsystems that make driving easier. The Spider may have a failsafe subsystem for when the user tries to use it outside its operating parameters but this hardly qualifies the device as a robot. I know that people can debate to death what the proper definition of robot is, but I doubt you will find anyone in academia who will classify this device as a robot. Even the promotional video attached to the Engadget story does not refer to the vehicle as a robot. It seems to me that today just about any remote-controlled device is called a robot and that is just sad.

Experimental robot OFRO to guard school children in South Korea

Last summer, OFRO provided security services during the World Cup in Germany. Now, the same robot is scheduled to enter a trial phase as a school guard protecting children from predators. The robot guard equipped with cameras will patrol the premises of a Seoul middle school looking out for outsiders who want to harm the children. The robot developed bu DU Robo, will be able to notify the teachers if it detects someone trying to seduce a student. OFRO can follow preprogrammed routes or be remote-controlled. The robot is not cheap as its estimated cost is $100,000 so don't expect to see it mass deployed any time soon. I wonder how good the robot's abilities to detect bad behavior are; it would be sad if parents start getting arrested because the robot saw them talking to children. There must be a huge shortage of teaching staff in Korea or a huge problem with child predators for this application to be thought off.

[Via Mail & Guardian online]

Microsoft unveils coffee-table computer with haptic interface

Today Microsoft unveiled Surface, a new computer system in the form of a coffee table and a 30-inch touch screen allowing for complex interaction with users. Computers with touch screens are nothing new since such devices have been around for many years. The important innovations are the 5 build in cameras that observe the world (can be used to read bar codes on objects) and the haptic interface which allows it to respond to more than one touch at a time. The latter makes it possible for the user to do things such as finger painting. In addition, multiple users collaborating on a project could work simultaneously such as for example manipulating two different photographs at the same time. The new device will cost somewhere between $5,000 and $10,000 so initially it is not meant for home use. In fact, Surface will make its first appearance in T-mobile stores and properties owned by Starwood Hotels & Resorts Worldwide Inc. and Harrah's Entertainment Inc.

The best way to appreciate the system is to watch it in action. So, here is a demonstration video that I found on YouTube (there are many more videos there including Microsoft's promotional videos,)

I really like the planned restaurant application. If you have ever had the pleasure of going out with a large group of people, then you probably know how hard it is to split the bill at the end. The Surface computer will make such a task trivial since people will be able to order food and drinks, then split the bill by setting down a card or a room key and dragging their menu items onto the card.

The only thing I wonder about is how is Microsoft's technology related to Jeff Han's multi-touch screen. For those who don't know, Han founded Perceptive Pixel Inc. to commercialize a touch screen he developed at the NYU Courant Institude of Mathematical Sciences. Check out this video of him showing off his creation,

Microsoft's product is clearly similar and so I am wondering if they licensed the technology from Han's company or if the just copied it. Coming to think of it, Apple's upcoming iPhone will also be boasting very similar technology. So, either Han is going to be a very wealthy guy after today or just another story of how the big companies screwed the little guy. In time, we will know.

Tyzx's DeepSea G2 3D vision system

As computers are becoming faster every year and computer vision research is making huge progress on new algorithms on a daily basis, 3D vision systems are starting to take their place in consumer applications. For example, stereo vision can be used in cars to detect other cars on the road or pedestrians; such information can be used to aid drivers in collision avoidance. However, such real-time applications require 3D vision systems that exhibit robustness to a broad range of lighting conditions, efficient processing, low power consumption and ease of programming. Tyzx's second generation stereo camera DeepSea G2 in conjunction with the DeepSea development platform promise to deliver an embedded stereo-vision system that satisfy these requirements.

The DeepSea G2 stereo-vision system consists of two CMOS imaging sensors, a specially developed application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a digital signal processing (DSP) processor and a Linux driven PowerPC in a small form package. The system perfrorms correlation-based stereo processing including background subtraction at 30 frames per second for high resolution images at 512x380 pixels resolution.

Tyzx says that the DeepSea G2 is being successfully used for applications ranging from people tracking in interactive displays to obstacle detection and avoidance in urban driving. The system sells for a pricey $4,995 for low volumes which I guess means that you have to purchase a large number of the units or else be prepared for an even steeper price. This is definitely not a device meant for hobby roboticists. Then again, there is no stereo vision rig available for amateur robotics which might just be a market worth looking into.

India's plan to use robots to assist astronauts on Moon mission

Well, I saw on Slashdot an article announcing India's plans to utilize a legged robot to assist astronauts during their scheduled manned Lunar mission in 2011. You can read the news article here but basically a small team of researchers at the Indian Institute of Technology Kanpur have created a two legged robot with camera and laser sensors for the bargain price of $50,000. The project is known as SmartNav. At any rate, the same people announced that the Indian Space Research Organisation (ISRO) has shown some interest in using a robot to assist astronauts in data gathering during the Chandrayan-II Moon Mission in 2011. The team admits that the current model will require substantial improvements in order to become space certified in addition to moving from a two-legged platform to a four-legged one for robustness. I wonder why not use a wheeled robot considering how successful NASA has been with such a configuration. Anyways, I would say that many people are overly excited about this announcement but I would suggest caution. The current prototype is hardly space certified and the team knows that it will take time and money to make it so. So the actual cost for a robot that would fly would be much higher than the bargain price quoted in the article. Additionally, there is no commitment by the ISRO to actually use or even fund this project; at least, not yet. So, at the time there are no plans to use a legged robot during the Lunar mission. If they do decide to use such a robot then they better get working on it right away because there are only 4 years left until lift-off and in the world of robotics this is not necessarily a lot of time.

Bayesian networks made easy with Netica

Inference using Bayesian Networks (BNs) and Influence diagrams is well study in academia and has a myriad of applications in industry. The only problem is that working with Bayesian Networks and keeping up with the latest developments is next to impossible. Norsys Software Corporation makes Netica the most widely used software package for working with Bayesian Networks and Influence diagrams.

A Bayes net is a model. It reflects the states of some part of a world that is being modeled and it describes how those states are related by probabilities. The model might be of your house, or your car, your body, your community, an ecosystem, a stock-market, etc. Absolutely anything can be modeled by a Bayes net. All the possible states of the model represent all the possible worlds that can exist, that is, all the possible ways that the parts or states can be configured. The car engine can be running normally or giving trouble. It's tires can be inflated or flat. Your body can be sick or healthy, and so on.

Although a basic knowledge of BNs is required in order to use the software, Netica’s graphical user interface allows one to specify and solve networks with ease. The software is constantly under development and new releases supporting recent work on BNs is always added. The software is used heavily in academia because of its robustness and large number of available features. Norsys also publishes APIs for low lever programming making it a suitable research platform. The company has also been very successful marketing Netica to the industry with many companies in engineering, biotech, aerospace, finance, insurance, mining and defense to name a few.


Netica is a rather expensive piece of software with a price tag of $585 for commercial use and $285 for educational or personal use. However, a free and full-featured version of Netica is available for download; the free version is limited to smaller networks.

Fire Scout robotic helicopter soon to join the U.S. army

Northrop Grumman Corporation is under contract to develop an autonomous flying helicopter for military use and they have now announced the first successful tests of the machine’s engine bringing the helicopter one step closer to production. The MQ-8B Fire Scout Unmanned Aerial Vehicle (UAV) is being developed under the umbrella of the U.S. Army's Future Combat Systems (FCS) project since its selection in August, 2003. According to the manufacturer,

The Fire Scout will be a key element of the Army's tactical intelligence, surveillance, reconnaissance and targeting architecture, providing real-time imagery, data collection and dissemination at the brigade level.

It should be noted that this machine is not remote controlled but truly autonomous with the ability to sustain flight for up to 8 hours if lightly equipped or up to 5 hours with a full weapons load. The Register is reporting that the U.S. Navy is planning to deploy up to 200 Fire Scouts less than a year from now.

I am not surprised that the military is interested in autonomous flying vehicles but I am troubled by the machine’s ability to carry weaponry. Even if flying is fully autonomous, I wonder if a human operator would have to verify a target before it is fired upon. If not, then it won’t be long after the Fire Scout’s deployment in a war zone before we start reading in the news about accidentally firing at civilians or friendly troops.

Ugobe announces that Pleo maybe available for pre-order in a month

In an email sent to investors and early adopters, Ugobe co-founder Caleb Chung announced that the packaging (seen in the above photo) for Pleo is now complete and that the robot is nearing a market release.

So, your burning question is when will Pleo arrive? As promised, we’ll announce an exclusive opportunity for members of the Pleo community mailing list to secure the first available Pleos so stay tuned to your email inbox next month.

So, if you want to be informed then sign up for updates here.

In addition to the sneak peek of Pleo’s packaging, Caleb also announced that in response to community feedback, the robot will come with a rechargeable and replaceable battery. I can’t really imagine a useful robot that would require its users to run down to RadioShack every few hours to buy new batteries. So, this is a good move for Ugobe but I wonder if this will result in a slightly more expensive Pleo. I guess we won’t know until the robot is finally available for purchase.

Stay tuned!

Google working on the 3D mapping of cities

News is that Google has licensed technology used by Stanford’s Stanley robot car winner of the 2005 DARPA Grand Challenge that they plan to use in order to create 3D maps of cities. The new data will be used to augment the maps currently available at maps.google.com and are supposed to rival the 3D maps published by Microsoft on their online maps service Virtual Earth. Apparently, Stanford researcher Sebastian Thrun will also work part-time for Google as part of the deal. There was rumors last March that Thrun’s stealth startup Vutool was being acquired by Google although no official information was made available at the time. Vutool had undertaken the task of mapping cities from street level using a fleet of cars equipped with a number of cameras and laser sensors. Other colleagues of Thrun’s in Europe are also working in creating 3D maps of cities in a similar way; see for example our previous coverage of the SmartTek project.

An artificial intelligence system for helping people with dementia

Older people with cognitive disabilities have a hard time completing everyday tasks that healthy people take for granted. This is especially true for people suffering from Alzheimer’s disease and other forms of dementia. Researchers from Dundee University in Scotland and the Universities of Toronto and Waterloo in Canada are working on automated artificial intelligence systems for helping such people in completing the everyday task of hand washing.

The team recently published an award winning paper at the 5th International Conference on Computer Vision Systems; the paper explains the need for such automated assistive systems,

Older adults living with cognitive disabilities (such as Alzheimer’s disease or other forms of dementia) have difficulty completing activities of daily living (ADLs), and are usually assisted by a human caregiver who prompts them when necessary. The dependence on a caregiver is difficult for the patient, and can lead to feelings of anger and helplessness, particularly for private ADLs such as using the washroom. Computerized cognitive assistive technologies (CATs) are devices that may have the potential to allow this elderly population to complete such ADLs more independently by non-invasively monitoring the users during the task, providing guidance or assistance when necessary.

The team led by Dr. Jesse Hoey has developed a vision-based system that monitors an individual during the hand washing process and verbally prompts when assistance is necessary. In addition, the system displays a demonstration video to guide the patient if the audio prompts are not effective. The system works in real-time and utilizes the framework of Markov Decision Processes (MDPs) to monitor the patient’s progress and it can automatically adapt to the patient’s level of awareness and dementia level.

The team has developed and evaluated the system using controlled experiments with actors in preparation for clinical trials which are currently taking place in Toronto. The following video shows one of the actor trials. The person intentionally fails to follow the proper sequence of steps during hand washing, and the system prompts him with the correct action. On the left-most part of the video, you can see a 3rd person view of the scene and on the right-most part you can see the view from the system’s camera including the tracking of the actor’s hands and towel. In the middle, you see information about the variables that the system monitors and have to do with the patient’s levels of responsiveness, progress and awareness.

The system is general enough that the team hopes to extend its use for other tasks such as tooth brushing and toileting. Upon successful completion of the clinical trials, the researchers hope that they might be able to commercialize their system to help the more than 5 million people in the U.S. alone suffering from Alzheimer’s disease; in fact, the Alzheimer’s association tells us that every 72 seconds a new person develops the disease.

The Robot Hall of Fame 2007 inductees

Carnegie Mellon’s Robot Hall of Fame announced earlier this past week the four inductees for 2007. As always, the lucky robots are a mix from science fiction and real-life. The honorees will be joining their brothers Shakey, ASIMO, Mars Pathfinder Sojourner Rover, R2-D2, HAL9000, Gort and C3PO and many others in the Robot Hall of Fame established in 2003 by the School of Computer Science at Carnegie Mellon University. So, here is the list of the 2007 Robot Hall of Fame inductees.

NavLab 5: Before Stanford’s Stanley autonomously navigated through the desert, researchers at Carnegie Mellon had already developed a series of autonomous cars that were able to drive at legal speeds on everyday roads and highways. NavLab 5 was the 1990 Pontiac Trans Sport that completed a trek named No Hands Across America driving autonomously more than 98% of the way across the U.S. from coast-to-coast.

Lego Mindstorms NXT: Lego was about to give up on the Mindstorms set when they got smart and asked the community for help updating it. The result was the Mindstorms NXT platform boasting a new powerful computer and sensors along with open source software that welcomed a large community of amateur roboticists who couldn’t wait to hack the new robots. The Mindstorms made robotics accessible to a large audience and it deserved to be honored in the Robot Hall of Fame.

Raibert Hopper: The Raibert Hopper is the one legged hoping robot developed in the 1980s at the Leg Laboratory at Carnegie Mellon University. The robot is named after its designer Marc Raibert who foresaw that feature robots would have to rely on the principle of dynamic balance if they were ever to match human locomotion abilities. The lessons learned with the Hopper proved central for biped, quadruped and even hexapod running.

Lt. Cmdr. Data: Born out of Gene Roddenberry’s imagination, Data was the android member of the crew on the starship Enterprise in Star Trek: The Next Generation TV series. Data boasting a positronic brain was engaged in a continuous search for its humanity while traveling among the stars and upholding the prime directive. Data is single-handedly the ultimate robot dreamed by every robotics researcher. We are still far from creating such a wonderful device but with rapid advances in algorithms and electronics, Data may actually become a reality within our lifetime. The question is whether such an android would be used for good or evil. Let’s hope it will be the former.

CoroWare brings cheap mobile robot to market

CoroWare announced the CoroBot mobile robot platform for amateur roboticists and academic researchers. CoroBot is another effort to bring an affordable and upgradeable robotic platform to market. The four-wheeled robot’s brain is a standard PC running at 3.2 GHz boasting 512MB of RAM and a 20GB hard drive. The basic CoroBot configuration comes with a web-camera capable of 640x480 pixels resolution. A more advanced model comes equipped with a robotic arm boasting 4DOF and an 8 ounce payload. The robot’s payload is maxed at 5lbs which in my opinion is rather disappointing. The cost for the basic model is $2499. Adding the robot arm increases the CoroBot’s price to $3499. The robot supports Microsoft Windows XP and Linux while drivers for the Microsoft Robotics Studio and the open source Player/Stage are also provided.

In general, the CoroBot seems like a better choice than the competing White Box Robotics 9-series PC-BOT which is priced at $2,995 for the basic unit and $4,995 for the more advanced model. On the other hand, the robot is still a bit too pricey considering that it is nothing more than a PC with a web-camera on wheels. If you are looking for a basic robot for an introduction to robotics then iRobot’s Create platform is probably a better choice considering that it costs only $299.99 for the premium development package (a basic model is only $129.99).

The iRobot Create Challenge

After DARPA's Grand and Urban Challenges and their European counterparts, iRobot decided to team up with Tom’s Hardware Guide to create a similar but low budget event centered on their Create robot platform. The Create Challenge is open to the amateur roboticist which is the target audience for the Create platform in the first place. The goal of the challenge is to

to find the coolest, most impressive robot that can be built with the Create platform.

There is a $5,000 prize for the contest winner and iRobot will also supply a few robots in the form of scholarships for those who can’t afford to buy one and want to enter the contest. There are only 15 such scholarships available so if you are interested then you better hurry up and apply before all are gone.

You can sign up for the contest at Instructables.com.

Read the Press Release about the Create Challenge.

Another Rodney Brooks interview

ZDNet’s Candace Lombardi has interviewed MIT professor and well known robotics scientist Rodney Brooks on the heels of the RoboBusiness 2007 event that is currently held in Boston. If you have read another Brooks interview in the last couple of years then you probably have read this one since Candace asks the same questions once again. The interview covers Brooks’ past and present and towards the end he talks a bit about current trends in consumer robotics while he avoids making any large predictions about the future. If you need a refresher on Brooks’ views of robotics then you can read the entire interview titled The robots are coming at the ZDNet website.

Academic humor: Chicken chicken chicken

Who says that academics can’t be funny? Take for example the following video from the 2007 American Association for the Advancement of Science (AAAS) humor session showing Doug Zongker presenting his seminal paper Chicken Chicken Chicken: Chicken Chicken (pdf). At first, I didn’t think it would be that funny but I found myself laughing outloud in no time. If you have had to sit through a few days of dry scientific talks then you will appreciate the humor in this video. Enjoy!

AMD demonstrates the first true quad-core CPU with the Phenom processor

Earlier today, AMD announced and demonstrated their newest microprocessor design boasting four cores on a single chip. The new CPU is named the Phenom processor and it differs from Intel’s Core 2 Duo which also has four cores in that the latter simply stuffs two dual core CPUs together; AMD has developed a single chip with all cores on it. The company said that the Phenom will be available later this year. At the same time, AMD demonstrated their upcoming eight-core CPUs codenamed FASN8.

The new CPUs promise better energy management and faster performance not only because of the additional cores but also due to more efficiently routing data to them.

AMD Phenom processors will be uniquely designed to facilitate intelligent uses of energy and system resources that are reliable, virtualization-ready and energy efficient, driving optimum performance-per-watt. All AMD Phenom processors will feature resources like an integrated DDR2 memory controller, HyperTransport™ technology links, and 128-bit Floating Point Units, for improved speed and performance in floating point calculations.

With the true quad-core design offered by the upcoming AMD Phenom processors, cores communicate on the die rather than through a front side bus external to the processor – a bottleneck inherent in other products that are packaging two dual-core chips to form quad-core processors. Additionally, AMD’s Direct Connect Architecture on-chip ensures that all four cores have optimum access to the integrated memory controller and integrated HyperTransport links, so that performance scales well with the number of cores. This design is also highlighted by a unique shared L3 cache for quicker data access and Socket AM2 and Socket AM2+ infrastructure compatibility to enable a seamless upgrade path.

I have said before that multi-core CPUs are great for much of artificial intelligence since many algorithms are easily parallelizable. In general, however, even though there are announcements from Intel and AMD almost every few months about new multi-core processors, I don’t hear as much about programming languages and compilers that will allow programmers to take full advantage of them. I get the feeling that soon we are going to have extremely powerful and cheap parallel computers and no easy way of programming them missing out on the hardware engineers’ innovations.

DARPA makes the first cuts for the Urban Challenge

Last Friday, the Defense Advanced Research Projects Agency (DARPA) announced the 53 teams out of a total 89 teams that qualify for the next round of the Urban Challenge. The teams selected will now be visited by DARPA personnel who will conduct additional tests to evaluate the abilities of the robots in preparation for the semi-finals in October and finals in November, 2007.

This evaluation covers a subset of the abilities robots will require to complete the Urban Challenge course, including merging into traffic, navigating traffic circles, negotiating busy intersections, and avoiding obstacles.

After the evaluation is completed, DARPA will make further reduce the number of teams to 30. These teams will be allowed to participate in the National Qualification Event (NQE) in October 21-31, 2007. DARPA will announce the location of the NQE and the final competition on August 10, 2007.

You can read the complete list of teams that have qualified so far here (pdf document.) In case, you are wondering, the usual suspects including the teams from Stanford and CMU that finished in the top spots during DARPA’s Grand Challenge have both qualified. Other notable teams that made the cut are from Berkley, Caltech and Princeton. Interestingly, there are also international teams from Mexico, Germany, France and Canada in the mix.

DARPA will be giving away $3.5 million in prize money to the top 3 teams to finish the course in less than 6 hours; remember that only the winning team in the Grand Challenged received a monetary award last year.

Marvin Minsky interview in Discover magazine

Last January, Discover magazine interviewed Marvin Minksy, MIT professor and one of the fathers of Artificial Intelligence as the co-organizer of the 1956 Dartmouth Summer Research conference that established AI.

In the interview, Minsky outlines his new theory for constructing an artificial brain as described in his new book, The Emotion Machine. He also does not miss out on the opportunity to beat up on neuroscientists and modern AI methods such as statistical techniques and neural networks while he expresses his fondness for commonsense reasoning and more specifically reasoning by analogy.

I don't see neuroscience as serious. What they have are nutty little theories, and they do elaborate experiments to confirm them and don't know what to do if they don't work.

In addition, he expresses his dissatisfaction for the lack of funding for traditional AI and the focus of government agencies to finance more practical applications of AI.

Funders want practical applications. There is no respect for basic science. In the 1960s General Electric had a great research laboratory; Bell Telephone's lab was legendary. I worked there one summer, and they said they wouldn't work on anything that would take less than 40 years to execute. CBS Laboratories, Stanford Research Lab—there were many great laboratories in the country, and there are none now.

Read the rest of the interview at the Discover magazine website.

Photo of Marvin Minksy is copyright Donna Coveney/MIT

Intelligent cars can achieve improved fuel efficiency

A recent study published in the Transportation Research Part C: Emerging Technologies journal compared the fuel efficiency between the standard combustion engine, hybrid and a telematics enabled car to show that adding intelligent control to a vehicle improved fuel efficiency by a large percentage. The study was conducted by researchers from the Department of Mechanical and Manufacturing Engineering at the University of Melbourne, Australia.

Using telematics, i.e., traffic information collected using a sensor network, a car with a standard combustion engine and intelligent control that utilized the received information was as efficient as a hybrid car while it outperformed a standard car by as much as 20% in fuel efficiency during urban driving.

The fuel economy of the optimal hybrid is found to have an average of 20% improvement relative to the baseline vehicle across three different urban drive cycles. Feedforward information about traffic flow supplied by telematics capability is then used to develop alternative driving cycles firstly under the assumption there are no constraints on the intelligent vehicle’s path, and then taking into account in the presence of ‘un-intelligent’ vehicles on the road. It is observed that with telematic capability, the fuel economy improvements equal that achievable with a hybrid configuration with as little as 7 s traffic look-ahead capability, and can be as great as 33% improvement relative to the un-intelligent baseline drivetrain.

Such an intelligent vehicle could be cheaper to produce compared to a hybrid because some cities already collect telematic information for the intelligent control of traffic lights.

Communication between a fleet of vehicles has been utilized in Automated Highway Systems previously (e.g. PATH) in order to improve the overall behaviour of a platoon of vehicles in response to changing traffic conditions. Thus it is assumed for the intelligent vehicle in this paper that there exists a sensor network potentially incorporating inter-vehicle communication, radar and laser technologies that can be used to convey information about the surrounding traffic. This traffic preview information can then be used to adjust the vehicle’s instantaneous velocity, whilst arriving at the destination at he same time as an un-equipped vehicle.

I think that this is a very interesting idea. I would prefer an intelligent system that does not take control of the car driving process but works under the hood to make driving more efficient and safe. Most of the automated highway driving systems that are being tested assume that all cars on the road are intelligent and in control; people in this cars are just passengers and not drivers. Such a system might result in more efficient use of highway real estate and higher fuel efficiency but most people own cars because they like to drive. I believe that future cars will employ more efficient engines and intelligent control systems that will work with human drivers in an unobtrusive way mostly monitoring the driver’s attention and taking control when an important event such as a red light or jaywalking pedestrian is missed.

Article: Fuel economy improvements for urban driving: Hybrid vs. intelligent vehicles

Robotics summer schools are great for young graduate students

In recent years, there has been a number of well focused summer schools designed to provide an in depth introduction of a robotics or artificial intelligence topic for young and senior researchers. These summer schools often have one-week duration and include both lectures by experts and hands on experience. This summer for example, the Portuguese Association for Artificial Intelligence will be organizing a summer school in Collective Robotics.

The summer school "EAIA07: Collective Robotics" provides an introduction and overview into Collective Robotics methods and issues with a particular focus on the biologically inspired swarm-robotics. There will be hands-on ateliers with real and simulated robots. It is primarily intended for young researchers from industry, PhD students or postdoc researchers investigating Robotics, Swarm Intelligence, Auto-organization, Multi-Agent Systems, Evolutionary Robotics, Social Insects, and related issues.

In the past, there has been much success with summer schools on Simultaneous Localization and Mapping (SLAM.) Additionally, for the last 4 years, the IEEE Robotics and Automation Society (RAS) and the International Foundation of Robotics Research (IFRR) have co-sponsored some very interesting summer schools on Human-Robot Interaction, Robot Design, Haptic Interaction and Learning. Notably, the first of these schools resulted in establishing the foundations for the emerging field of Human-Robot Interaction (HRI) which for the last couple of years has also had its own ACM/IEEE-sponsored international conference.

For more information about the upcoming schools go to the following websites,

Collective Robotics Summer School
Learning Summer School

Gigapan: high-resolution panoramas using consumer-level digital cameras

Researchers from Carnegie Mellon University and NASA Ames have created an affordable system for creating high resolution panoramic images using consumer-level digital cameras. Consumer cameras can capture images of 5-10 megapixels resolution and limited field of view. Now, NASA is trying to commercialize the imaging technology that they used on the Mars Exploration Rovers to create astonishing panoramic images of the planet. The resulting technology called Gigapan can create wide angle images of several gigapixels resolution by seamlessly stitching together several low resolution photos. The photos are taken from different positions and zoom creating a panorama which can be zoomed-in to reveal additional details of a scene and basically allow for its virtual exploration.

Gigapan is a low cost base system with pan and tilt capabilities that works with a large number of consumer-level digital cameras. The hardware is expected to be priced at around $200 making it affordable for amateur photographers as well as professionals.

Many digital cameras come equipped with software that can construct panoramas but they are not of the high quality wide angle photos that Gigapan can generate. One has to visit the product page and look at the panorama of the Grand Canyon and other scenes to realize the wealth of information and flawless stitching of the images to appreciate the system. Looking at the examples on the project website, I noticed that the system is capable of handling moving objects in the images avoiding ghosting effects but not entirely; in the example panorama of the Waffle at Burning Man 2006, no ghosting is visible but in the example of the Guatemala Market the ghosting effect is definitely prevalent in several places. Unfortunately, there is no current technology that can adequately deal with this problem in highly dynamic scenes although I am certain some computer vision researcher is working hard on solving this problem.

There are many other commercial and open source computer vision projects concerned with the problem of constructing high resolution panoramas from a collection of low resolution photos. One good example that does not require additional hardware similar to Gigapan is the Autostitch software developed at the University of British Columbia; a similar open source version of this software is called Autopano-SIFT. Also, Microsoft is developing Photosynth which works by creating panoramas from images collected over the Internet.

Gigapan is part of the Global Connection Project that aims to use the power of images to connect people around the world.

Microsoft’s Robotics Studio under fire by the academic community

I just came across a blog post by Gregory Dudek who is a professor and director of the McGill Research Center for Intelligent Machines in Montreal, Canada, about Microsoft’s efforts to promote their robot development platform at the recently concluded International Conference on Robotics and Automation (ICRA 2007.) Dudek reports that Microsoft’s Tandy Trower gave a long presentation promoting MSRS but his efforts were met with criticism by the audience that is not in favor of a closed and proprietary software development environment. According to Dudek,

Herman Bruyninckx from K.U Leuven (and head of Euron) gave a very spirited rebuttal to Trower and argued that every time Microsoft enters a new market the range of alternative solutions becomes much smaller and more polarized. He was quite provocative and got a very strong level of support from the audience. At present, there are several alternative open source robotics toolkits and packages around.

In other words, the research community at the moment is in favor of supporting Open Source initiatives for robot programming, namely the Player/Stage platform that in recent years has served the community very well.

I finally downloaded a copy of the latest version of MSRS and I am looking forward to playing around with it. I am mostly interested in the physically-based simulation environment because the one available in Player/State is rather primitive. My first impression from trying out some of the demos provided by the Microsoft team is that the system might be more complicated than it has to be. On the other hand, complexity is inevitable when one tries to build a general system that would work with a large variety of hardware platforms that have little in common with each other.

I am curious to see how this latest reincarnation of the Open Source versus Microsoft battle will play out.

Speech recognition research video from 1968

An anonymous tipster sent me a video describing speech recognition research at Stanford vintage 1968. The video is interesting in more than one ways. First of all, the video is a great introduction to the basic difficulties behind developing a robust speech recognition system, including finding word boundaries in continuous speech and disambiguating between phonetically similar words and phrases. Second, the video shows the real-time operation of a speech recognition system 40 years ago that is used to control a robot arm operating in a blocks world. Speech recognition systems today are still faced with the same difficulties recognizing continuous speech even though faster computers and statistical models have improved their overall robustness. Finally, I really like the production values behind this video with a great description of the science involved and all the auxiliary characters demonstrating the fundamental issues making speech recognition challenging such as the difficulty in distinguishing between the two sentences, "Ice cream" and "I scream." I wonder if this is the type of video that the people at AAAI are thinking about with the AI Video contest that we mentioned yesterday.

Anyways, here is the video,

AAAI video competition announced

The Association for the Advancement of Artificial Intelligence (AAAI) announced a new video competition to take place during the organization’s 2007 conference to be held in Vancouver, Canada. The contest is about videos showcasing AI projects.

AAAI is pleased to announce the launch of the AAAI-07 Video Competition. The goal of this competition is to communicate to the world how much fun AI (research and application) is and, in particular, to document exciting research and applications using artificial intelligence. The rules are simple: Compose a short video about an exciting AI project, and narrate it in a way that makes your video accessible to a broad online audience. We strongly encourage student participation. This is your chance to make a cool online video about your AI research and/or application, and get a ton of attention!

Video format: Either 1 minute (max) "short video" or a 5 minute (max) "long video", with narration in English (or English subtitles). Consider combining screen shots, interviews, and video of a system in action. Make the video self-contained, so that newcomers to AI can understand and learn from it. We encourage a good sense of humor, though we will only accept submissions with serious AI content. Your video might cover contemporary research, or document seminal AI research in the past. Creativity is encouraged!

In case you are wondering, other than the prestige of winning the first ever AAAI video contest, there is a pot of $3,000 to be shared among the winners and also cool little trophies inspired by SRI’s historic Shakey robot.

More info on the video contest at http://www.aivideo.org.

Mowgli: a jumping robot

Researchers from the School of Interdisciplinary Information Studies at the University of Tokyo and the Intelligent Systems Research Institute (AIST) in Japan are working on a new prototype bipedal jumping robot. The robot nicknamed Mowgli (likely a tribute to the well known book character) utilizes an artificial musculoskeletal system with pneumatic muscles to perform explosive motion jumping as high as 50% the robot’s height and landing smoothly. The musculoskeletal system is inspired by biological systems; as the team explains in a recent paper, animals which are good at jumping and running have tapered legs with a small foot mass and moment of inertia.

Mowgli stands less than 1 meter tall (0.9 meters to be exact) and weighs just 3Kgrs. As mentioned earlier, the robot is bipedal with 6 pneumatic muscles on each leg for 6 degrees-of-freedom. The robot is controlled by an off-board PC running a real-time Operating System (OS) while the compressed air for the muscles and electrical power are also supplied by off-board systems. The team has demonstrated the robustness of their robot performing experiments in simulation and with the real robot humping on the ground and onto a chair. In fact, here is the video of Mowgli jumping from the ground and landing on top of a computer chair without human intervention,

Video is copyright the School of Interdisciplinary Information Studies, University of Tokyo, and the Intelligent Systems Research Institute, Japan.