We wrote about robotic road trains last year, and somewhat remarkably, the research project that’s trying to make them happen is progressing nicely with help fro companies like Volvo. Cars drafting closely behind trucks in road trains save up to 40% in fuel consumption, thereby saving money and the environment at the same time. But the biggest advantage of being in a road train is that you can just stop paying attention to the road and do something else while your car drives itself.

It’s true that your life is in the hands of the system, and not in your hands, but while the current perception is that that makes things more dangerous, it really should be the exact opposite. With your car doing all the hard work for you, accidents are less likely.

Of course, the system is only as safe as the lead driver, so many different technologies are being employed to make sure that the person driving the truck is sober, qualified, and paying attention. This includes a breathalyzer and a fancy infrared camera system for vision tracking to make sure that the driver is paying attention to the road.

Other especially exciting bits from the video include mentions of ‘several years from now’ for road trains themselves, and the suggestion that technology to allow cars to drive themselves in stop and go traffic jams might be just around the corner. As we’ve discussed before, the technology (adaptive full stop cruise control and lane keeping) is already here and in some cars, we just have to catch up in terms of people (and lawyers) being comfortable with it.

VIA [ Wired ]

Artificial intelligence systems are good at tackling problems that can be solved using brute force, like chess… All the computer has to do is calculate out every possible permutation of moves and pick the best one. They’re also pretty good at games like poker, where even with incomplete information, a computer can make a move that is statistically ‘best.’ And lastly, they’re good at making decisions far more quickly than a human.

When you combine all of these separate characteristics into one game, things get exponentially more complex, but also much more like real life. And this is why people are trying to teach computers how to play StarCraft, at a level where they can compete with even the best human players.

UC Santa Cruz hosted the 2010 StarCraft AI Competition, which put AI programs through a series of different StarCraft testing scenarios to determine the most effective AI system at micromanagement, small scale combat, tech limited games, and of course full gameplay. The video above shows a bunch of highlights; especially notable is the absolutely brutal use of mutalisks by the eventual AI winner, UC Berkeley’s Overmind.

The last clip in the highlight video shows an AI taking on a world class human player, who wins handily. It’s only a matter of two or three years before humans have no chance against programs like these, however… And the reason (I think) is quite straightforward: the computer can micromanage every single unit it owns, on every part of the map, at the same time. A human can’t. Once the AI reaches a competent level of strategy and unit use (it’s not there yet), we’re screwed, because the AI can just launch multiple simultaneous micromanaged attacks.

There are lots more videos of the different AI programs competing against each other on YouTube here, and you can download in-game replays at the link below.

[ StarCraft AI Competition ] VIA [ New Scientist ]

The robotics journal Autonomous Robots has its own blog, which is intended to take the hardcore robot news from the journal and make it a bit more reader friendly. They also link back to the journal articles, should you need a little of that hardcore techy info. Yeah baby. Anyway, looking back through some of their posts, I found these vids of LittleDog exhibiting some new learning behaviors.

While it’s best to develop a robot that’s capable of adapting to new situations, after the robot explores the world for a bit, new situations (in the general sense) stop showing up as often. So ideally, you want a robot that’s able to recognize a situation that it’s already experienced, and apply those past experiences without having to start figuring out what to do from scratch. To this end, Martin Stolle and Christopher Atkeson are developing a system that allows a robot like LittleDog to first recognize features that it’s seen before (like walls or gaps), and then access a library containing a sequence of actions that it has successfully applied to solve similar situations in the past. LittleDog can then alter those actions to apply them to the current situation rather than attempting to compute an entirely new action sequence. It’s faster, more efficient, and I imagine more successful since LittleDog is building on past experiences that worked as opposed to trying something entirely new.

After the jump, another LittleDog video, just because.

[ Finding and transferring policies using stored behaviors ] VIA [ Autonomous Robots Blog ]

Thanks Sabine!

Yesterday, we posted an update on PR2′s for sale status, and included a video from an informal contest sponsored by Willow Garage founder Scott Hassan, inviting PR2 beta teams to submit cool / funny / useful PR2 videos to be judged by himself, his wife, and his kids. The above video (from Ping Chuan Wang, Stephen Miller, Mario Fritz, Trevor Darrell, Pieter Abbeel at UC Berkeley) took first place and a cool $5000, which is way more than any person has ever been paid for folding two socks together.

Second place went to Bosch, for their PR2 mailman, whose name appears to be “Alan:”

UPenn took third place with their one robot band:

Really, each of these videos is deserving of its own post, but I wouldn’t do that to you… You can check out the other six (there’s also PR2 StrongBot, posted yesterday) after the jump.

Towel Folding: Social Network Edition, from UC Berkeley:

The PR2 picks up a towel from an unknown configuration and proceeds to fold it. When it is unsure which corner would be best to grasp, it polls Twitter for advice. In return, once it has finished the procedure, it posts a photo of the folded towel.

To interact with the PR2 and assist it in future demos, follow now at http://twitter.com/berkeleypr2

This PR2 is named Brett, maybe?

Book Bot, from MIT:

This one seems to be named M&M.

Self Portrait, from Bosch:

We wrote a custom controller, a GCode parser and some infrastructure to allow the PR2 to draw arbitrary images. Of course, the first thing it drew was itself.

PR2 Red light green light, by JSK:

PR2 Painter, from KU Leuven

PR2 with a beard. Epic.

PR2 Head Tracking, from UPenn:

There’s one more video called PR2 Checkout, from Stanford, but it gave me a headache so I’m just going to link you to it instead.

VIA [ Willow Garage ]

Almost exactly one year ago, Nexi took part in a rather silly (but reasonably effective) demonstration of a robotic role on a US Navy vessel. I guess the Navy was impressed, ’cause they’re now the proud owners of their very own Nexi, which they’ve named Octavia. What, exactly, is the Navy going to do with an MDS robot? Well, duh, they’re going to teach it magic tricks:

Your tax money at work. No, seriously, I wholeheartedly approve of my tax money being used on robot magic. The Navy, though, isn’t just about the tricks… They’re hoping to use Octavia to explore how humans and robots interact, with the aim of minimizing the amount of time that humans spend dealing with a robotic interface, and maximizing the amount of information that can be communicated. Octavia specifically is good for things like this because of her intensely expressive (and only mildly uncanny) face and head.

If you’re lonely and like the look of Octavia, you can buy one for yourself at the link below.

[ MDS Robots ] and [ Xitome* ] VIA [ Gizmag ]

*This website seems to have a virus associated with it, according to my virus scan. To go there anyway, click here.

To be fair, some of PR2′s Poolshark programming team look to be pretty terrible at pool, but that doesn’t make it any less impressive that in only five days, PR2 learned how to hold and shoot a pool cue, recognize ball locations, select the best shot, and then sink it. If you’re wondering what this robot can’t do, the answer seems to be nothing (besides using stairs and round door knobs).

Willow Garage has two more of these week long ‘hackathons’ planned this month, which will include teaching PR2 how to push a cart (meh) and fetching drinks from a fridge (yes please). A robot that can play pool and fetch me beer? Hellooooo new best friend.

[ Willow Garage ]

Human World Cup Soccer (that’s football, to most of you) is fun to watch (even if it is on at 4am here in California), but as far as evolution goes, we humans have pretty much peaked. Really, the only thing about the game that evolves reliably from cup to cup is the ball (and that’s not always a good thing). Robots, on the other hand, have no such limitations. Carnegie Mellon’s CMDragons small-size robotic soccer team have taken another step towards robot domination by teaching their small size soccer bots the physics of ball movement. The demonstration below pits a robot that knows physics against a robot that doesn’t; keep in mind that the robots are entirely autonomous, controlled by a computer that watches the action on an overhead camera:

Without modeling the physics of the ball, the computer just tries to position the robots on the ball without taking the movement of the ball into account. A physical model allows the computer to move the robot predicatively, greatly improving dribbling skill:

Call me crazy, but these robots look to be demonstrating the same basic skills as talented human soccer players: win the ball, keep your body between the ball and your opponent, and then get an angle, get around him, and shoot.

Here’s a video of the CMU team competing in 2008, without the physics based planning software… Two years ago is a long, long time in the world of robots, but even so, the passing and shooting is impressive:

And if soccer’s not your thing, the same basic skills make the robots highly effective at playing minigolf:

I guess now we know what soccer robots do for fun in their spare time.

RoboCup 2010 starts in Singapore on June 19.

[ Stefan Zickler ] AND [ CMU Robot Soccer ] VIA [ Robots.net ]

Well, I didn’t get one, but ten eleven research groups are getting their very own PR2s to mess around with as part of the PR2 Beta Program. Out of 78 submissions, Willow Garage was forced to (somehow) choose only ten eleven, and those lucky sods will get a PR2 for a couple years. The goal is to generate a whole bunch of new ROS libraries that other robots running ROS will be able to use to do everything from the laundry to the dishes. And, you know, other stuff.

There are going to be some fascinating developments coming out of this program. And since it’s all open source, if you have a robot capable of running ROS, you may actually directly benefit from what these groups come up with. Summaries of all of the accepted proposals, after the jump.

Albert-Ludwigs-Universität Freiburg with the proposal TidyUpRobot

The University of Freiburg will teach the PR2 to tidy-up tasks like clearing a table, while working on difficult underlying capabilities like understanding how drawers and refrigerators open, and how to recognize different types of objects.

Bosch with the proposal Developing the Personal Robotics Market: Enabling New Applications Through Novel Sensors and Shared Autonomy

Bosch will bring their expertise in manufacturing, sensing technologies and consumer products. In addition to software contributions, Bosch will be making robotic sensors available to members of the PR2 Beta Program, including a limited number of “skins” that will give PR2 the ability to feel its environment.

Georgia Institute of Technology with the proposal Assistive Mobile Manipulation for Older Adults at Home

The Healthcare Robotics Lab at Georgia Tech will be placing the PR2 in an “Aware Home” to study how robots can help with homecare and creative assistive capabilities for older adults. Their research includes creating easier ways for people to interact with robots and enabling robots to interact with everyday objects like drawers, lamps, and light switches.

Katholieke Universiteit Leuven with the proposal Unified Framework for Task Specification, Control and Coordination for Mobile Manipulation

KU Leuven in Belgium is a key player in the open-source robotics community. As one of the founding institutions for the Orocos Project, they will be improving the tools and libraries used to program robots. They will also be working on getting the PR2 and people to perform a task together, like carrying objects through a crowded environment.

MIT CSAIL with the proposal Mobile Manipulation in Human-Centered Environments

The MIT CSAIL group will use the PR2 to study the key capabilities needed by robots that operate in human-centered environments, such as safe navigation, interaction with humans via natural language, and planning for complex goals. Their work will allow robots to build the maps they need in order to move around in buildings as large as MIT’s 11-story Stata Center.

Stanford University with the proposal STAIR on PR2

PR1 was developed in Kenneth Salisbury’s Lab at Stanford and ROS was developed from the STAIR (Stanford AI Robot) Project. The team will work on several applications, which include navigating a building, identifying objects, retrieving items scattered about a building, and clearing a table after a meal.

Technische Universität München with the proposal CRAM: Cognitive Robot Abstract Machine

TUM will research giving PR2 the artificial intelligence skills and 3D perception to reason about what it is doing while it performs various kitchen tasks. These combined improvements will help PR2 perform more complicated tasks like setting a table, emptying a dishwasher, and other kitchen-related tasks.

University of California, Berkeley with the proposal PR 2 Beta Program: A Platform for Personal Robotics

The PR2 is now known as the “Towel-Folding Robot”, thanks to the impressive efforts of Pieter Abbeel’s lab at Berkeley. In two short months, they were able to get PR2 to fold fifty towels in a row. Berkeley will extend on its work with the PR2 folding towels and tackle the much more difficult challenge of doing laundry, from dirty laundry piles to neatly folded clothes. In addition, Berkeley’s team is interested in assembly and manufacturing tasks through learning by demonstration.

University of Pennsylvania, GRASP Laboratory with the proposal PR2GRASP: From Perception and Reasoning to Grasping

The GRASP Lab proposal aims to tackle some of the challenges facing household robotics. These challenges include tracking people and planning for navigation in dynamic environments, and transferring handheld objects between robots and humans. Their contributions will include giving PR2 a tool belt to change its gripper on the fly, helping it track and navigate around people, and performing difficult two-arm tasks like opening spring-loaded doors.

University of Southern California with the proposal Persistent
and Persuasive Personal Robots (P^3R): Towards Networked, Mobile,
Assistive Robotics

USC has already demonstrated teaching the PR2 basic motor skills so it can adapt to different situations, such as the motions necessary to pour liquid into a cup. They will continue to expand on this work in imitation learning and building and refining skill libraries, while also doing research in human-robot interaction and tools so the robot can use sensors more accurately over time.

University of Tokyo, JSK Robotics Laboratory with the proposal
Autonomous Motion Planning for Daily Tasks in Human Environments using Collaborating Robots

The JSK Lab’s goal is to see robots safely and autonomously perform daily, human-like tasks such as retrieving objects and cleaning up. They’ll also be working on getting PR2 to work together with other robots.

[ Willow Garage ]

BotJunkie got invited to the official dedication of the Volkswagen Automotive Innovation Laboratory at Stanford yesterday. You remember VAIL, right? They’re the people who taught this car to park itself.

Anyway, this was our first invitation to an event involving a major head of state, so after we got through the Secret Service checkpoints and the camera sniffing dogs, we got to see the dedication ceremony and ribbon cutting by Angela Merkel, the Chancellor of Germany:

And here’s what was behind the ribbon:

As you can see, Shelley got a sexy new paint job for the occasion. At this point, she’s more or less all ready to take on the Pike’s Peak course autonomously… Next up are some high speed just-for-fun-and-breaking-the-robotic-speed-record runs out in the desert, and the duplication of a portion of the course on some flat ground without any cliffs to see how she does in the turns at full speed. The only major hardware concern is that Shelley currently does not have any obstacle avoidance sensors. So, if some hapless hiker decides to trek up Pike’s Peak at the same time as the robot does, there’s the risk of some hiker bits getting stuck to the front grille and ruining the aerodynamics of the car. One option is some form of telemonitoring, but they’re still working through ideas.

After the VIPs and their armored SUV convoy left, we had a chance to play around a little bit, and snagged ourselves a ride in Junior 3, VAIL’s self-parking Volkswagen Passat:

While we were sitting in the car for this demo, the way it’s ultimately designed to work is that you get out of the car, and tell it to go park itself using the iPhone app. When you want it back, hit the return button on your phone, and it comes back to you. It’s that easy. For more on the technology behind this, check out our post from last year on Junior 3′s autonomous parking capability.

I don’t know! Here it is! Watch it! Be happy!

You’re welcome!

[ RoMeLa ]