MIT is teaching Little Dog new tricks in navigation.

Sometimes robot videos just make me laugh. CSAIL at MIT has been working with Boston Dynamics’ robot Little Dog, helping it navigate rough terrain in novel ways. The scrappy quadruped can dynamically shifts its weight on two legs at a time, helping it climb slopes and stairs, and generally get around.And as soon as Little Dog gets where it’s going, it promptly flops down on its belly much like a real canine. The careful steps followed by exhausted collapse gets me every time. Check out the video from BotJunkie below, and look towards the end (1:44) to see for yourself.

Little Dog’s journey is part of Phase 2 of DARPA’s Learning Locomotion Program. As those who read our War 2.0 story know, a larger version of the robot, aptly named Big Dog, is being bred to work as a mule for soldiers in the field. That bot can haul loads and keep walking even after a hefty kick (see its video below). The navigating and stepping routines that CSAIL teaches Little Dog are going to be directly portable over to Big Dog.

Having a robot make its way on a rough path isn’t new. We’ve seen how Asimo navigated terrain using an overheard camera and symbolic shapes. CSAIL’s robotic gardeners also operated under self-guidance. Little Dog is doing much the same: an overhead camera helps the bot choose it’s next steps carefully while it makes decisions and maintains its balance dynamically.

The leaps and jumps it makes may not seem impressive, but there’s definitely potential there. Accurately moving in the air and planning where to land are some of the more difficult tasks walkers have to make on a routine basis. If you want to be furthered impressed by Little Dog’s stepping prowess, check out this video from last year. It looks like CSAIL decided to give the robots some Shaolin training. Very cool.

Pattern recognition, problem solving, and decision making are all necessary when walking, and they are all tasks humans excel at. In fact, these are perhaps the only skills where humans still defeat computers every time. The fact that CSAIL is helping Boston Dynamics and DARPA develop a robot that learns as it walks is important stuff. Robotic learning is still in its infancy, and as with human infants, walking is just the first of many goals. With a strong foundation in problem solving, all learning machines need to exceed human intelligence is time. That is if they don’t get tired and flop out first.

Wait...now go!..No! Wait...wait...go, go, go!

ASIMO’s success may prove to be a launching point for walking robots. Bipedal bots are some of the most challenging designs in the robotics world. Not only do you have to worry about balance and locomotion, they have to find a safe place to put their feet. Which is why ASIMO’s performance was so exemplary. Carnegie Mellon’s new foot-stepping software allows the bot to plan out complex and dynamic paths in its environment. It is constantly asking itself, what’s the best way to get where I’m going?

Of course, ASIMO wasn’t alone. Carnegie Mellon’s Robotics Institute developed a generalized foot stepping program for all legged robots, not just Honda’s. The Linux based H7, the HRP-2 series (remember the sexy 4C that we covered early?), and even the “little dog” (we featured the “big dog” in our war robots story) had a chance to find their way through difficult terrain and elevations. Each robot had its own course, and its own videos, but only ASIMO seemed poised to step out in the real world.

It’s only a hop, skip, and a jump from point A to B

As you can see in the video, the man behind the robot is Professor James Kuffner. His approach to planning and testing the foot stepping program is simple and elegant. It all comes down to weighted costs. Think about when you want to cross difficult terrain. You first look at obstacles and determine which of them would be harder to cross than others. Such cost-analysis helps you find the easiest path to take. That sort of decision making is what Kuffner is helping robots to do as well.

Kuffner allows the bots to rate the terrain based on different factors: angles, safety, roughness, stability, and bumpiness. Combine all those factors together and the robot can choose which objects to avoid, and which to head towards. When I watched the video, I was impressed that ASIMO could dynamically evaluate the obstacles quickly enough to dodge two blades spinning at different rates. That’s really a great accomplishment. Of course, it would be a lot more impressive if the “obstacles” weren’t actually paper cut outs.

Difficult terrain appears red, easier terrain appears blue or green. A robot can then plot its steps along the easiest path.

It would also be more impressive if ASIMO couldn’t rely on the overheard camera. That eye in the sky provides 2D information that walking robots probably wouldn’t have access to in the real world. At this first level test, Carnegie Mellon has excelled, but getting robots to use their 3D vision to create a 2D map of their area and navigate it…that would be a wonderful next step. Some of the 3D stereo image recognition work has already been done with a navigation robot in Germany.

Pardon me sir, but your robot is on my foot…

The really promising thing about Kuffner’s weighted costs approach to footstep planning is that robots can be programmed to assign a cost to almost anything. People are probably the most important. As readers have pointed out, Sushi Chef robots, flexpickers, and baseball robots are all a little scary when you think about how much power and speed they possess. By assigning an infinite cost to human-robot collisions, you could insure that robots do everything in their power to avoid running into people. In a crowded office building, on a street, or even in your home, bipedal robots will need this type of software to help them navigate around rapidly shifting humans in their environment.

But the benefits of cost-analysis don’t stop there. You could get bipedal robots to value moving walkways (those conveyor belts at the airports), escalators, or other helpful terrain that humans use all the time. By adding in hand movements, you might see robots that can climb ladders or the sides of buildings. Such search and rescue robots are just one option. In fact, Carnegie Mellon’s research is really opening up the possibilities for bipedal robots. I mean, the whole reason they’re designed with legs in the first place is so that they can take advantage of human-like versatility of movement. Kuffner’s group is helping walking robots access this potential.

I’m not sure I agree , however, with Kuffner’s opinion that robots will be seen in the home as soon as the price tag drops down below $50,000. As cool and versatile as humanoid robots may be, they aren’t the only option out there. The Roomba, after all, is the top selling house-hold robot and its far from humanoid. With advances in automobile navigation, the standard robot in the house may be a car.

But at some point, humanoid robots will likely become much more present in our daily lives. Maybe as workers, maybe just as novelty. Either way, the navigational abilities that Carnegie Mellon has developed will be invaluable. I’m anxious to see what Kuffner and the robotics institute has in store for ASIMO and his robot friends next. Hopefully they’ll move away from the 2D world of Frogger and jump into using on board cameras. Either way, I’m confident that these scientists will help robots put their best foot forward.

Autonoumous City Explorer (ACE) navigated Munich by the friendly finger-pointing of humans.

So what makes this lost robot so unique? Unlike its tween counterpart, ACE isn’t just picked up and pushed in the right direction. The bot is roughly human sized and packed with instruments to help it detect and query humans for help. First, there are cameras and image recognition software geared towards finding people and moving ACE towards them. When a human is found, an audio message is played while virtual lips move on a screen. If the human is friendly (aren’t we all?) he or she uses a touch-screen to indicate it’s willingness to give directions. The human will then point in the direction ACE should go.

And ACE follows your finger! That’s really kind of cool. By using multiple cameras and more image software, ACE is able to build a 3D model of the human and where he or she is actually pointing to. Compared to a toddler, this isn’t a remarkable skill, but following visual cues is really difficult for most robots out there. ACE also prompts its helper human to use the touch-screen to suggest a given path, but in the end, it can rely on just the pointing.

ACE’s point and go navigation avoids the hassle of interpreting humanity’s subjective audible directions. “It’s over there, behind that building, then you sort of take a right and keep going for a while.” —that’s a message fit to make a robot pull its hair out. ACE just has to query enough people (38 for its maiden voyage) and it will build up a developing map of its surroundings. Simple and effective, the ACE system allowed the robot to travel little more than 1km in 5 hours. Not a speedy journey, but very promising.

Just wait ’til your robot comes home!

While Tweenbot is definitely the cute one, and ACE is light years ahead in technology, both lost bots illustrate how humanity and robots are being taught to work together. According to ACE’s mandate, it seeks to develop new cognitive abilities to make robots more useful in a human-rich environment. Likewise, Tweenbot was an experiment in humanity’s generosity (read here: kindness to something foreign). As robots leave the factory and warehouse (sorry KIVA), they will have to accommodate the needs of, and rely on the input from humans.

The ACE experiment showed that the robot-human interaction could be successful, but it wasn’t very efficient. It also demonstrated human willingness to interact with the artificial, but would this trend continue once the robot’s novelty wore off? The secret is making robots more life-like (what’s more human than wandering around asking for directions) and making humans more robot-aware. Human-like robots may be destined to take a larger part in our society, but only if we decide to accept them. Experiments like ACE and Tweenbot show that the potential exists, we just have to find it. Luckily, we have people (and robots) willing to point us in the right direction.