During the Fukushima crisis in Japan, the lack of Japanese robots that were available to help out was notable. There was some question as to why Honda didn’t just send ASIMO (arguably one of the most sophisticated and capable humanoid robots in existence) to help out. The simple answer is that ASIMO wouldn’t be able to handle that kind (or any kind) of extreme environment. The robot was never intended to be a disaster mitigation robot; it was designed to work in offices, specifically the kind of offices that have not experienced an earthquake, explosion, alien invasion, sharknado, or other messy event. Honda is clearly aware of ASIMO’s limitations in tackling these kinds of situations, and that’s probably why (as we reported two years ago) the company has been developing a new version of ASIMO that is specifically designed for disasters.
At the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) this week, Honda engineers presented a pair of papers on research they’re doing with disaster-response humanoid robots. The researchers report that they’ve been focused on complex tasks such as gait transitions and ladder climbing. It was nice seeing that their ASIMO-based experimental humanoid is already doing some very impressive things.
The two papers that Honda presented at IROS weren’t actually about this new robot at all: the first was on “Dynamic Gait Transition Between Bipedal and Quadrupedal Locomotion,” and the second was “Robust Vertical Ladder Climbing and Transitioning Between Ladder and Catwalk for Humanoid Robots.”
It’s clear that the company is still working on this new platform and isn’t ready to officially introduce it to the world. The robot [pictured right] doesn’t even have a name. Honda refers to it as “experimental humanoid robot.”
Here are some specs from the papers, compared to the specs of ASIMO:
It looks like there are some sensors in the head that we can’t quite make out, but most notably, it doesn’t appear to be tethered for power, and that giant white box is either batteries, some kind of fuel cell, a miniature fusion reactor, or a gerbil on a wheel. It’s evident that the new robot is very different from ASIMO, although we have to assume that there’s a significant amount of ASIMO DNA that would go into any new humanoid robot from Honda. And for the record, we did ask the researchers to tell us more about everything (or anything, really), but they referred us to Honda PR, which means that nobody is going to tell us anything.
Now let’s take a look at the research that the Honda engineers presented, which provides a few clues about what the company might be planning for its robot.
In the DARPA Robotics Challenge Trials, robots had to climb a very steep set of stairs. That was pretty hard, and in the DRC Finals, the stairs were shorter and not steep at all. Honda has no time for stairs, and has their robot climbing a vertical ladder, including transitioning onto it and off of it from a narrow catwalk. We’ve never seen a robot demonstrate this behavior before. It’s difficult to do because ladders are slippery, and small slips, rotations, or other positional errors and motions that vary from planned trajectories can result in a fall.
The control method that Honda has developed makes continuous real-time estimates of the posture of the robot and adjusts subsequent contact positions to compensate for any errors, resulting in “robust multiple rung vertical ladder climbing and bidirectional transitioning from a ladder to a catwalk.” Apparently the robot is able to ascend fairly briskly, and Honda said it is working on even faster and more efficient ladder climbing gaits, including “trotting” and “pacing.” I assume these are terms for climbing gaits that use fewer than two steps per rung.
The DRC also highlighted the advantage of multimodal locomotion: robots that weren’t restricted to walking, but could also drive on wheels or switch to a quadrupedal mode whenever it was necessary or advantageous to do so. Honda is also exploring this idea, except they’re trying to do it without any compromising of their humanoid form. Essentially, they’re making a robotic ape.
Unlike other humanoid robots that can transition to quadrupeds, Honda’s robot can do so dynamically, without maintaining a static center of gravity. This makes it very quick, able to autonomously go from standing to all fours in about 2 seconds. To do that, it relies on neat trick: it spins a pair of flywheels in the torso. How cool is that? It can get back up again just as quickly, and move continuously at about 0.5 km/h the entire time. The balancing software is an extension of the algorithms that keep ASIMO upright, and the researchers hope to develop autonomous planning software that will allow the robot to locomote dynamically through arbitrary environments.
From the current state of this research, we wouldn’t be surprised if Honda had started in on this project almost immediately after Fukushima. They’re irritatingly good at keeping secrets over there, so we’re feeling pretty lucky that we even got to see this much. Hopefully, we’ll be getting sporadic updates from now on, since it seems like Honda has decided to let its researchers present at conferences. We’re not at all sure what the endpoint is, or when we might see a robot that could be called complete, but it’s fantastic that Honda is actively working on disaster robotics, and we’re very much looking forward to learning more. Eventually.