Photorealistic, Reliable 3D Mapping For Robots

An Anonymous Coward writes: "Hans Moravec at Carnegie Mellon University has updated his DARPA-funded MARS program page , including new info about the possibily of having photo-realistic 3D mapping for robots in the near future. "Our machines will navigate employing a dense 3D awareness of their surroundings, be tolerant of route surprises, and be easily placed by ordinary workers in entirely new routes or work areas. The long-elusive combination of easy installation and reliability should greatly expand cost-effective niches for mobile robots, and make possible a growing market that can itself sustain further development...We expect our new data to bring us further towards photorealism, and more importantly extremely reliable 3D maps." Check out all the slides and movies at the bottom of the page."

Comments

Computer Vision's Goal

[fireboy1919]

I am a student studying computer vision. It has long been known that CMU, and especially the famous Dr. Takao Kanade have been interested in autonomous vehicles - that and stereo vision have been the major focus of that school. They're very good, but this is really not just an event at their school. Its more like a community (the vision robotics community) effort. Even Slashdot has posted a few of the other contributions.

HOWEVER, its still a long time coming. In addition, space projects and industry require much more precision and accuracy than academia can offer alone. Keep in mind that CMU already did the "No hands across America" project, where there cars "drove" (they controlled breaks and gas) 99% of the way across the United States autonomously. That was a while ago - so was their total virtual environment mapping dome. Have you seen any autonomous cars for sale? What about 3d videos that you can see from every point of view?

We still have a long way to go.

I saw something similair to this at MIT

[hooded1]

While on a tour of the MIT AI labs i was shown a project they were working on, which has similair components to this. Essentially what they were doing was, creating a system in which they could point to a part of the room (with their hand) and vocally tell the computer to project some image there. The project is called Hal The Next Generation Intelligent Room

MIT is doing something similar

[Binary Tree]

MIT City Scanning Project: Fully Automated Model Acquisition in Urban Areas

Very interesting

Classic example of SMPA

[mtDNA]

Moravec's approach is a classic example of the SMPA (sense-model-plan-act) approach to mobile robotics. A lot of people think this is a dead end - not least among them Rodney Brooks, who advocates what is called the behavior-based approach. Behavior-based robotics basically relies on integrating several independently operating reflexes into a robot, which is much more lifelike. A nifty intermediate approach is taken by Ronald Arkin, who seems a little more pragmatic (and less dogmatic).

You can read some superficial information about all of these guys (and others) in the book Robo sapiens.

A review of Robo sapiens can be found here.

Re:Classic example of SMPA

[isaac_akira]

Moravec's approach is a classic example of the SMPA (sense-model-plan-act) approach to mobile robotics.

No it isn't. It's just a vision processing system that creates an internal model of the 3d environment around it. Once you build that model, you can do whatever type of reactive behavior or goal-oriented planning you wish.

Re:Classic example of SMPA

[Animats]

Sigh.

Early thinking (60s-70s) really was to build a detailed model of the world, grind it down to simple primitives, and run a logic-based planner on it. That had a terrible time dealing with uncertainty and required a very regular world.

Moravec introduced the idea of "certainty grids", which are probablistic occupancy maps. Originally, he used this as a means of getting useful data from ultrasonic rangefinders, which are very low resolution devices with slow data rates. (I've built a robot that works that way myself, and you really can get maps with more resolution than the sonar beam by taking enough samples as the robot moves.) As enough compute power became available, he moved to laser rangefinders (better resolution, but clunky rotating mirrors) and finally to passive stereo imaging.

What you get out of systems like this is a map of the neighborhood showing what's open space and what isn't. This is a good input to a repulsive-field type path planner. There's no need to extract a "primal sketch" or do any object recognition just to accomplish navigation using this approach. It works quite well; the CMU Navlab vehicles have been cruising around offroad on this technology for years now. The Denning guard robots used this technology with sonars.

Extracting range data from stereo imagery was Moravec's thesis topic in the 1960s. It took a mainframe computer 20 minutes per frame back then. Now it can be done in real time. There's commercial software for this. Two cameras are good; three cameras are better. It's actually not that hard; it's basically done by convolution. It's not done by edge recognition any more. Convolution is computationally expensive, but simple. We finally have enough compute power to do this stuff.

I've commented on Brooks' work previously, so I won't say any more about that now.

One question...

[mduell]

bring us further towards photorealism, and more importantly extremely reliable 3D maps

Ok, my only question... have they used it for porn yet? (they always seem to use technology first ;p)

Other posts you are likely to see:
1. I want a 3D map of Natalie Portman!
2. Can they make a beowulf of these?
3. Im gonna pour hot grits on the robot!

Ok, I just needed to get that out of my system.

Mark Duell

Problem with Brook's work...

[Anonymous Coward]

The problem I've always had with Brooks' work (besides the whole "Subsumption Arch is the ONE TRUE WAY" ego trip) is that he always tried to claim some kind of biological relevance where THERE WAS NONE.

I am currently a graduating senior in both biology and computer science, and am very interested in the integration of both, in the areas of neuroethology and biomimetics. Most of the people doing biomimetic robotics (i.e. robots quite strictly based on biological systems, theories, data, and constraints) don't like Brooks for that reason. He created robots, vaguely insect like - they used no real biomechanical data, neural control data, etc., and seemed to suggest that there was some real insect structural and behavioral aspects to them. I also think he's somewhat of a playboy, going from so-called "insect-like" robots (Genghis, Attila, etc.) to his media/attention/funding grabbing monstrosity COG (not that I think $$$ going to AI research is bad in any way), which he makes claims with hubristic abandon about its learning abilities.

I bet in a few years, when interest and funding dies, and he sees what a complex, deep hole he's dug himself into, he'll think of something new to grab headlines about...

BTW for those really interested in more "hard" work on biologically inspired control and networks, look at Eve Marder's page at Brandeis (which I don't have with me at the moment) and have a nice gander at

http://neuromechanics.cwru.edu

The Case Western Reserve University's new graduate program in neuro-mechanical systems. There, Dr. Roger Quinn, and many other researcher are working on some great biomemetic projects, including a robo-roach, and cricket. They use hard biological data to design these guys with. They also do significant work in neural basis of behavior, biomechanics, and neural-controlled prosthetics. I'm also plugging it because there is a 50-50 chance I will be attending the PhD program there next year in sunny, gorgeous Cleveland, OH!

Sincerely,
Kevin Christie
kwchri@wm.edu