Entries : Category [ Robotics ]
Articles about robots and robotics.
[Miscellaneous]  [Computers and Technology]  [Travel]  [Education]  [Hacks]  [Robotics]  [Science]  [Programming and Software]  [iPhone]  [Digital TV and Video]  [Intellectual Property & Copyright]  [Personal] 

16 January

CBC interview on our plans for multi-robot collaboration at sea

This morning I did an interview with Dave Bronstetter, one of the great Canadian radio broadcasters, and a familiar voice to Montrealers. The interview aired live on the radio show All in a Weekend on the Canadian Broadcast Corporation (CBC) station Radio One. The interview was quite pleasant and relaxed, and a marked change from the very first radio or TV spots I did a long time ago when I was very nervous.

We discussed the work in my lab on multi-robot mapping using a team of heterogeneous robots, some of which can swim underwater, some that can fly and a robot boat. While we have been working on multi-robot systems from over a decade, it is only recently that we have gotten into outdoor experiments with such a diverse set of different systems. By combining both swimming and flying, we gain a tremendous ability to map and understand a region of the shoreline.

I may have goofed a bit in understating the number of other groups working on multi-robot systems: there are lots. Before in the pre-interview we had discussed the number of groups using technologies like ours, which is small, and this is what I was thinking about during the interview.

One of the particular pleasures of the interview stemmed from the fact that Dave Bronstetter had interviewed my father many times, and remembered him well. We had a chance to talk about him a bit before and after the on-air segment. In addition, Dave's voice and manner was very familiar since I have listened to him on various shows over a long span of years. He's also served as television anchor man and host of the popular and long-running radio show Daybreak, and I think I may have spoken to him then.

You can listen to the MP3 recording here, or use one of the embedded players below.

Music Player Extended

Above embedded player requires Flash7 plugin, click to start.


Dave Bronstetter on air CBC
Dave Bronstetter on air CBC

I have been interested in multi-robot collaboration for several years, and have worked on it with various students. Some of the key themes have been: using multiple robots to help one another figuring out where they are by having them observe one another (the PhD work of Ioannis Rekleitis on collaborative localization), allowing robots to find one another even though they know nothing about the environment and cannot communicate (the MSc work of Nicholas Roy on rendezvous and "what to do when you're lost at the zoo"), and various problems on multi-robot search. Papers on some of this can be found at our lab's publications page at http://www.cim.mcgill.ca/~mrl

By Gregory Dudek at | Leave a comment |    
15 February

Tonight [updated below] was the second installment of the match on the TV show Jeopardy between past champions Jim Jennings and Brad Rutter, and the IBM computer system called "Watson." Watson is essentially a server room full of of IBM Power7 computers with thousands Processors. It uses a "community of experts" approach to answering questions.

Answering Jeopardy questions is impressive since the questions are phrased in complex ways that are not naturally suited to computer search. You can't put the typical Jeopardy question into a Google search and get a useful answer. Furthermore, even if and when you can use the question of get at useful data sources, such as a Wikipedia page, the information that comes back is rarely in a sufficiently simple form.


The server room that constitutes the IBM Jeopardy-bot is biggest than many people's living rooms and crammer with high-end gear. Watson uses 2800 Power7 CPUs which have at last cores each, so that's like alost 6000 normal "computers" just in terms of sheer processing power (notwithstanding the computer power for a Power7 core vs a standard desktop CPU).

Watson works by looking up words and phrases in its' large database of documents and then sorting the many hits into a top-2 list. One of the tricks it uses is to analyze past Jeopardy games and see the patterns in the questions. One of these, apparently, is that the first and last words in a question are more directly connected to the answer. This may explain one of Watson's mistakes on the first day: When the clue was "“Stylish elegance, or students who all graduated in the same year,” the correct response from a human (posed in the form of a question) was “What is class? ”Instead, Watson went first and said "What is chic?”-- note, however, that "chic" is a synonym for the first word-pair "stylish elegance".

Ken Jennings as observed that a key in Jeopardy is hitting the buzzer at the right moment, and that most expert players know the answers to most questions This implies that one reason Watson does so well relative to the humans is the timing of the buzzer-pushes, rather than the question answering itself: that's a bit of a disappointment.

Jennings was also told he looked "grim" and responded in a Washington Post interview as follows: "Believe me, I was enjoying (almost) every second. Getting beat on the buzzer is frustrating, but are you kidding? I AM PLAYING A PRIME-TIME GAME SHOW AGAINST A SUPER-ADVANCED ROBOT! This is the coolest thing I will every do in my life by a factor of a million. The future is here."

Watson is made up from Power 750 Express Servers, each with four 3.55 GHz POWER7 eight-core processors, for 32 cores per machine. With 90 machines in 10 racks, this makes for 2880 cores. The systems has 16 Terabytes of memory, and 4 Terabytes of clustered storage. For the Jeopardy game Watson is using SUSE Linux Enterprise Server. Each individual Power7 system has a performance of 218 Linpack gigaflops whereas the Deep Blue computer that played chess had a capacity of 11.4 Linpack gigaflops for the whole system.

Update after game 3

Watson finally won the tournament with a resounding lead over the human players. The final outcome did have an element of luck in it. Midway through the game Watson was in second place and if the game had ended there it would have lost, but the take-home message is that it could play Jeopardy extremely well and beat almost any person to might compete against it.

A big question is how much Watson's win depended on its ability to simply click the buzzer as soon as possible after it was allowed to do so. It was clear that in many cases the people playing knew the answer to a question, but were frustrated by their inability to click the buzzer as fast as Watson to get their answer in. Well, that's the nature of this particular game. Again, the real message is that we now have a large scale system that is able to answer very indirect questions automatically, Precisely how to stacks up against Brad Rutter or Ken Jennings in the various facts of Jeopardy saavy is probably not that critical, although it would help us better understand the limitations of Watson's abilities.

By Gregory Dudek at | Leave a comment |    
18 September

Today I gave a one-hour talk at the IEEE 9th International Symposium on Robotics and Sensor Environments (ROSE). I talked about our work in underwater robotics (as I often do) and specifically dealt with summarization, terrain identification and a tiny bit about human-robot interaction.

One of the components of the talk was work by my former student Philippe Giguere, now a professor and Laval University in Quebec City. He was in the audience which made that part of the presentation feel a bit weird to me -- here I was reminiscing about work he had done with me and almost sharing the odd inside joke with him while this audience was listening. It was a very nice audience and they asked a lot of good questions, which left me with a generally positive vibe. Not a bad way to start a Sunday morning, after all!

The key coauthors of the work I presented, in addition to Philippe, are Yogesh Girdhar and Junaed Sattar.

The summarization part of the talk is covered (at a very high level) in my TEDxMcGIll talk shown here.

By Gregory Dudek at | Leave a comment |    
13 November

I the last couple of weeks I became the parent of a new baby. Oddly. there are quite a few other parents. Also, our child has 6 legs, 3 eyes, and aluminum body.

We have recently constructed the latest robot in the Aqua family, which has a few new design innovations that can be back-ported into some of the older robots too.

By Gregory Dudek at | Leave a comment |    
02 October

I am at the "Breaking the Surface" meeting in Croatia. It's an interdisciplinary meeting that brings together several different communities interested in underwater technologies and specifically underwater robotics. This includes computer science people like me, engineers, archeologies, biologists, vendors, and government. It is a great chance to see who the solutions, and problems, that we have and offer intersect.

Today I spend time out on the water attempting to use a couple of different kinds of system and vehicle, including the VideoRay, a remotely operated vehicle (ROV) that is connected to the operator by a copper cable. As ROV's go, it is pretty economical with an entry price of just $6,000, but that's without vision, transponders or other goodies. A fully decked out unit can run as much as ten times as much or more, but that's just because fancy sensors cost big bucks, as well all know.

On Saturday I will present some of the work from out lab including the issues and characteristics of our robot and its locomotion mechanisms, the summarization work being done by my student

By Gregory Dudek at | Leave a comment |    
16 December

Two weeks ago the McGill daily had a long rambling article that implied -- in fact more than implied -- that robots from our lab were going to be deployed in Afghanistan. The subtitle of the article was "Robots partly built at McGill set for deployment in Afghanistan." When I saw this, it literally spoiled my dinner. On one hand, it is disturbing to see even a student-run newspaper run an article that included what I see as misleading, naive and potentially damaging content. On the other hand, perhaps I should have just shucked it off as the vagaries of the rumor mill -- much the way movie stars are advised to ignore the yammering insanity of the supermarket tabloids.

The article combined a variety of unrelated facts and a large dose of innuendo to reach a totally unsubstantiated and incorrect conclusion. Our lab has, in fact, focussed on the use of robotics for environmental assessment, specifically the observation of coral reefs, for almost a decade (although we have also thought more recently about applications like monitoring the port of Montreal, or search and rescue). I have often been pleased by the fact that we might be making a tiny step towards making the ocean environment better, at the very least by attracting publicity to the plight of the reefs. The key research we do is based on automated analysis of visual images to allow robots to better interact with humans in their immediate vicinity, for example by allowing human scuba divers to more efficiently program them.

The key elements of the "expose" in the McGill Daily article was that (1) our research grew out of prior research that had been funded partly by the US military/industrial complex, (2) that a McGill professor who quit 10 years ago subsequently worked for a while for a company that also does research funded by the US military, and that (3) some professors someplace at McGill might be getting military funding. Just wait until they hear the bad news about Santa Claus...

Like all science and engineering, our work is built upon the results of others, and one of the important progenitors of our hardware platform was a project that was funded a decade ago partly via DARPA -- the US defense advanced research projects agency. None of this DARPA long-ago funding came to me, my students or my lab, but since we used ideas or technologies from a DARPA project this seems to have been enough for the paranoid authors of the article to draw all sorts of ridiculous conclusions for their hurried little magnum opus. Now, the ill informed might still worry because DARPA was somehow involved in this line of research at some point. Such a worry could only be spawned if somebody failed to recognize the role DARPA has played in US science and technology policy and funding for the last 40 years. DARPA has also funded studies into how the neurons in the brain are inter-connected. Yes, is also the same DARPA that single handedly funded the development of the Internet (originally called DARPANet). It's also the agency that funded projects to build the driverless cars now being deployed by Google (in fact by the same people including Sebastian and Mike who won the DARPA-sponsored competition). Oh yes, and they have taken on the quest for synthetic blood and the revival of the US manufacturing sector. In short, if you try to eschew technologies that DARPA helped start, you better get your bear skin rug and book of earthworm recipes ready for the comfy torch-lit cave you plan to settle into with the other drooling luddites.

A related, and more profound issue, but one that was apparently too subtle for the simplistic shovel-ware that was this article is the following: as science and technology continues to mature and expand human capacity, what mechanisms can we envision as a society to channel the results away from harmful applications? This can't just be a matter of an edict, or one or a few scientists, or even nations, staying away from certain lines of research. Rather, it will require a global-scale agreement on how certain technologies should, or should not, be exploited. As a rule, however, we have done very poorly at such efforts and any successful approach will have to be very forward-looking, multi-national and subtle and will have to avoid the heavy-handed tactics that have consistently failed in the past.

The McGill Aqua2 robot hard at work
The "death-dealing" McGill Aqua2 robot hard at work on a coral reef
(Click to expand)

By Gregory Dudek at | Read (1) or Leave a comment |    
Prev  1   2   3   4   5   6   7   8   9   10   11   [12]   13   14   15   Next