Sarcos exoskeleton video

Utah's Sarcos revealed yesterday their latest prototype exoskeleton for allowing humans to perform tasks requiring strength above that supplied by their own biological muscles. Even though I believe exoskeletons are an important technology for improving the life of differently enabled people, Sarcos seems to be mostly targeting their exoskeleton to military use; however, I am willing to bet that the primary reason for this is that the funding for the technology is supplied by the US military specifically to develop technology for use in the battlefield. As you can tell from the video, the early models will mostly be useful in terms of allowing soldiers to perform heavy lifting that otherwise would require much effort. Later models are clearly designed for use in battle something that I find a rather sad use for a great technology. At any rate, below is the video showcasing the capabilities of the latest Sarcos exoskeleton prototype.

Thanks Gary for the link to the video.

Twenty-one: a robot for your home

The Japanese unveiled today another robot designed to help care for their quickly ageing population. Researchers from Waseda University, presented their new home service robot named Twenty-One. The new robot looks massive standing just less than 1.5 meters tall but making up for its lack of height with lots of width. It is also rather heavy weighing nearly 250 pounds. The robot has a humanoid upper body with two arms; the upper body is mounted on top of a wheeled base creating a more stable robotic platform compared to the current state-of-the-art in legged robots.

TWENDY-ONE is an advanced version of the WENDY robot (Waseda Engineering Designed　sYmbiont) which was developed in 1999 in Sugano Laboratory, Waseda University. WENDY equipped the passive impedance mechanism in each joint of its two arms. Passivity is a key technology which is required for robots to perform various tasks with human in daily life. WENDY also has the human mimetic dexterous hand and it is the first robot in the world, which could break an egg skillfully.

In order for the robot to safely co-exist with humans, researcher have equipped it with a large number of touch and force feedback sensors allowing to detect and react to collisions in order to prevent injuries. The researchers demonstrated the robot helping an elderly person get out of bed and preparing breakfast with him. I would not be surprised if the demo was carefully designed to show the robot operating perfectly but this is almost always the case in such situations. The Twenty-One robot team wants to commercialize their creation by 2015 which would be plenty of time for them to work out many details that would allow the robot to operate in more complex and unconstrained situations.

Twenty-one robot official website with more information and videos.

EMIEW 2 robot crashes during demonstration

Hitachi's second generation humanoid robot didn't do very well during a recent demonstration for the public when it crashed on a desk saved in the nick of time by one of the reporters present. Apparently, the robot was under remote control at the moment but because of an increase in wireless traffic during the demo the operator lost the ability to send commands to the robot; the end result was an embarrassing crash for EMIEW (and Hitachi's robotics engineers) which is designed to operate as an office assistant. This should be a lesson to the Hitachi team because they should have had at least some basic obstacle detection and avoidance running on board the robot so when communication with the remote servers was lost, the robot would have continued to operate safely; this can be a real concern for a service or home robot because it could mean the difference between falling down stairs or running over pets and babies as opposed to actually achieving its specified task.

CBC has a more detailed report on Hitachi's EMIEW 2 failed demonstration.

PS: If anyone has video of the event let me know.

Roboquad holiday robot gift

If you are looking for an affordable and exciting robot gift for this holiday season then consider WowWee's Roboquad spider-like robot. The robot costs close to $120 and it comes loaded with functions guaranteed to keep any kid entertained for a while. Alternatively, you can pay nearly $400 for a Pleo and hope that it ships in time for the holidays. WowWee has a good history developing toy robots that are fun to play with; let us not forget that a couple of years ago, their first robot Robosapien was the hottest holiday gift toy. Check out one of WowWee's Roboquad TV commercials showing off some of the robot's functionality.

Salaries for advanced computer science degree holders

Last month I mentioned that CNN published results of a survey regarding the salary of robotics engineers. Today, I was browsing Indeed.com and I noticed that they offer a feature that lets one query their large database of job postings to get an idea of the average salary for professionals. I thought I check out how the salaries for computer science PhDs compare depending on their field of expertise, i.e., machine learning versus computer vision versus robotics and so on. The image below (click it for a higher resolution version) shows the results,



As you can see from the graph shown above, computer vision and machine learning graduates get paid the highest average salary. I am a bit surprised that robotics engineers are paid much less on average although I could guess that the reason for this has to do with the fact that most of the jobs involve industrial robotics which probably means a job that has to do with recalibrating robot arms in assembly lines; probably a job that can be easily done by someone with an undergraduate engineering degree and not requiring an advanced graduate degree. In the results, I have also included the more general “computer science” term to get an idea of the average pay for people with only an undergraduate degree.

I can't say that the results shown above are the most accurate but it is clear that going to graduate school can increase one's salary substantially. Talking to friends who are responsible for hiring new graduates and from my own personal experience looking for work the last couple of years, I have to admit that the average salaries computed by Indeed Salary Search are rather accurate. It is unfortunate that the standard deviation is not also given but only the average. In general, I would say that exceptional people are probably paid much better than the average so if you decide to go to graduate school then it is up to you to make the effort necessary to make it worth something.

PS: Indeed explains how they compute the average salary as follows,

Indeed Salary Search is based on an index of salary information extracted from over 50 million job postings from thousands of unique sources over the last 12 months. Many job descriptions don't contain salary information, but there are enough that do to produce statistically significant median salaries for millions of keyword, job title and location combinations - in fact, most job searches you are likely to think of. As new jobs are added each day, the Indeed Salary Search index is automatically updated with fresh salary data, so the salary results are as up-to-date as they could possibly be.

Netflix progress prize winners

If you remember, roughly one year ago, Netflix announced a contest with a $1 million prize for anyone who can design an algorithm to improve the accuracy of their recommendation engine by %10. They published a large amount of data about movies and user preferences for anyone wishing to participate. Lots of teams signed up as you can see from the official leaderboard. A year later, no team has achieved a result good enough to claim the full prize; the best improvement is close to %8.5.

Out of all the participants, a team of engineers from AT&T has devised an algorithm outperforming all others with results good enough to claim Netflix's progress prize worth $50,000. The AT&T team codenamed BellKor consists of Bob Bell and Yehuda Koren from AT&T's Statistics and Information Visualization departments. The method used by the AT&T researchers is explained in a paper they have published on their website.

Netflix has said that they will incorporate the algorithm in their recommendation engine within a year. As far as I can tell, Netflix stands to make much more money than the $50K they just had to pay for this improvement. If you think about it, they managed to get an unbelievable amount of labor performed by thousands of highly skilled people (apparently nearly 27,000 teams registered) having paid only the salary equivalent for a single junior software engineer! This trick of running a competition promising a high pay-off to the winner while setting the bar for the final prize high enough that the chances of anyone winning are very low is probably the best trick for a company to pay next to nothing for a massive amount of skilled labor. Not to mention that %8.5 increase in recommendation accuracy will result in higher movie rentals or in other words a profit that very likely will surpass the prize money paid.

That said, the team's achievement is well worth the recognition that they are receiving considering the massive amount of data that any algorithm must crunch through in order to make good recommendations for movie buffs.

The service robot Markovito

Markovito is a service robot designed for delivering verbal messages and objects among people in a laboratory or office-like environment. The robot interacts with users using speech recognition. In the video below showing the robot in action, we can see that speech recognition still has a long way to go before it is reliable enough for use on consumer-oriented service robots; let me also remind you of an older post on How to Say No to a Robot.

Toddlers warm up to humanoid robot

Scientific American reports on recent work at the University of California, San Diego, designed to study the social bonding possible between toddlers and robots. The researchers deployed a humanoid robot (the now defunct Sony QRIO) in a class of toddlers during a 5-month period. The robot operated both autonomously and also under remote control. According to the article,

The tots began to increasingly interact with the robot and treat it more like a peer than an object during the first 11 sessions. The level of social activity increased dramatically when researchers added a new behavior to QRIO's repertoire: If a child touched the humanoid on its head, it would make a giggling noise.

The researchers used a teddy bear toy and a less interactive robot as controls; interestingly, the humanoid's appearance was not the only factor in gaining the children's attention.

For 15 sessions midway through the experiment, QRIO was programmed to repeatedly dance to the same song rather than interact with the kids. During these trials, the children became far less interested in the friendly automaton. For the final three sessions, however, QRIO could once again unleash its entire social arsenal.

It is very interesting how a machine that exhibits social behavior is accepted by children which in my opinion necessitates further study of Human-Robot Interaction (HRI). In fact, HRI is a newly established branch of robotics with its own conference; HRI was first established as a field of study 3 years ago during the 2004 RAS/IFRR summer school on Human-Robot Interaction.

Read Could Robots Become Your Toddler's New Best Friend?

Omni-hand NASA prototype for sale on eBay

Smart Machines reader Gary alerted me today that a piece of robot history has been placed for sale on eBay. I'm talking about the prototype hand known as Omni-Hand I and this is apparently a working system purchased by the eBay seller at a NASA surplus sale. According to the description on eBay (starting bid is set to $500),

This impressive early prototype demands an important place within robotics history as the first motorized dexterous robotic hand. It represents one of the early steps towards making robots more anthropomorphic. The Omni-Hand was designed and built in the early 1990s by robot pioneer Mark Rosheim with funding from NASA contracts NAS8-37638 and NAS8-38417 for NASA. Two prototypes were made. The first was a "test bed" whose features were then incorporated into this complete unit. Both had the same power and control system.

Also, check out the following video showing the Omni-Hand grasping an egg and light bulb,



Additional information: Omni-Hand I eBay auction.

[Thanks Gary]

Robot suitcase from Russia

Russian inventors want to sell you a robotic suitcase capable of following you around in crowded environments using a large variety of sensors. According to Russia InfoCenter,

A gyroscope, light sensitive detectors, ultrasound and infrared sensors help the smart suitcase bypass obstacles, to roll in conditions of an inclined surface, and to stop when stumbling upon the edges of staircases and balconies. The robot-suitcase’s accumulator charge is said to be enough for non-stop operation during 2 hours.

The robot will take advantage of a card mechanism possibly using RFID to detect when it has fallen too far behind its owner; in the latter case, the robot will trigger an alarm to alert its owner of the situation. The company expects that the robot will be able to carry as much as 30Kgr and it will come with a build in charger and all the necessary plugs and adapters so that the batteries can be recharged in more than 100 countries without extra hassle.

The robot suitcase is still in the development stages while the inventors hope to begin mass production in 2009; they expect to sell the suitcase for the rather steep price of over $1900.

The official robotic suitcase page can be found here (in Russian so you might have to use a translation tool to read it.)

Graphical Model Algorithms website

The researcher at UC Irvine have announced a new website focused on Graphical Model Algorithms, GraphModAlg@UCI.

Its purpose is to make available implementations of the algorithms our group develops for reasoning in graphical models. We are concerned with algorithms for solving reasoning problems over graphical models, which includes common tasks for belief, constraint, and mixed networks.

We provide various exact and approximate algorithms for Bayesian network tasks as well as constraint networks and integer programming problems. We also host a repository of example problem instances

Software implementations of a large number of algorithms are available including implementations for Bucket Elimination and Iterative Join Graph Propagation (IJGP) among many variants of these and exact inference algorithms. See the software page for more information about the algorithms and their implementations.

I think this site will be a great resource for those interested in Graphical Models.

Machine Learning summer school

The 2008 Machine Learning (ML) summer school is now open for students and researchers who want to register. This is the 10th ML summer school since the first one in 2002. The school will be held at the Australian National University at Kioloa.

Machine Learning is a foundational discipline of the Information Sciences, concerned with the design and development of algorithms and techniques that allow computers to "learn".

Topics will be covered in 9-10 lectures a 6 hours taught by world experts in their fields. The aim of the summer school is to cover the entire spectrum from theoretical foundations to practical applications. In addition, there will be practical "lab" sessions, where students will have the chance to implement methods for themselves.

This school is suitable for all levels, both for people without previous knowledge in Machine Learning, and those wishing to broaden their expertise in this area. It will allow the participants to get in touch with international experts in this field. Exchange of students, joint publications and joint projects will result because of this collaboration.

Material is directed both at outstanding participants without previous knowledge in machine learning, and at those wishing to broaden their expertise in the area; this includes PhD, Masters, and advanced undergraduate students, postdocs, academics, and IT professionals. The MLSS also provides an excellent opportunity for interaction with top researchers in a broad cross-section of machine learning disciplines.

More information about the 2008 ML summer school here; information about past schools can be found here.

Robot arm video

I have written before about the excellent cable-driven robot arm (Whole-Arm Manipulator) by Barrett Technology Inc. The company was demonstrating their newer generation robot arm during the IROS conference and I took a video of the WAM in action. If you listen carefully, you can hear the Barrett engineer describe the benefits of the new arm and its cost (close to $200K fully loaded.)

ActivMedia Seekur mobile robot

ActivMedia Robotics was the largest exhibitor at the recently held IROS conference. They had demos of many of their mobile robots including the very large, four-wheeled Seekur. The robot is designed for indoor and outdoor operation but as you can tell from the video below, you might want to be careful when using it on carpet. Seekur is a large robot weighing 350Kgrs and can go on operating for 7 hours on a single battery charge.

More information about Seekur can be found here including more videos of the robot in action.

CMU wins the Urban Challenge

According to a Popular Mechanics article, the Tartan racing team from CMU won DARPA's Urban Challenge leaving competitor Stanford in second place. Third place went to the Virginia Tech team. Remember that in the past, CMU and Stanford finished in the first two places during the 2nd Grand Challenge with Stanford taking home first prize. CMU will collect $2 million dollars from DARPA for the win and Stanford another $1 million while Virginia Tech will enjoy $500 thousand.

The cars were able to drive autonomously for 6 hours sharing the road with human drivers. DARPA's scoring of the event is not publicly known but apparently CMU finished the course 20 minutes faster than Stanford.

Don't expect to find cars autonomously driving around town any time soon though. Most of the sensor packs used in the event are too large and clumsy to be mounted on a regular vehicle; not to mention that further study is necessary in terms of safety. Would the cars still be able to perform in rainy weather and icy roads? Even for military use there is still lots of work to be done. I bet most of these vehicles would fail badly if one of the many sensors were to be damaged. It would be easy for a soldier to render one of the vehicles useless by shooting at the sensor array.

Still, the ease with which teams completed the challenge is a testament to the great advances in AI and robotics during the last decade. I expect that in another 10 years autonomous machines will be an integral part of our daily life similar to the way the Internet had penetrated our daily routine. It is great to be involved in robotics and artificial intelligence during such an exciting period of time.

The final results and additional information about the Urban Challenge are available at the official website here.