Robotics is tricky, and human-like robots especially so, admits one of the field’s leaders. But each small step is one giant leap for robotkind.
Korea Advanced Institute of Science and Technology (KAIST) Professor of Mechanical Engineering Jun-Ho Oh, head of 2015’s DARPA Robotics Challenge-winning team, is intimately familiar with the difficulties of getting human-like robots to do anything but the most basic, repetitive tasks.
Oh said Team KAIST learned two main things: robots can do certain tasks, but the team had to do much more to make it work in a real situation.
KAIST won the notoriously difficult set of events with a little help from their DRC-HUBO robot’s unique adaptive design. Its ability to both walk as a bipedal and roll by getting into a kneeling position (using wheels on its knees and casters on its feet) “proved key to its victory” according to an analysis piece by IEEE Spectrum.
The three-year competition was inspired by the 2011 Fukushima disaster and the role robots might have in responding to a meltdown or similarly serious situation. In the finals, Oh’s team beat out the Florida Institute for Human and Machine Cognition and Carnegie Mellon University Tartan Rescue by completing all eight events in 44 minutes 28 seconds to claim the US$2 million award.
These included driving a car, egress from that car, door opening, valve turning, cutting through a wall with a tool, a surprise task, navigating debris and climbing stairs.
The difficulty even in stable, bipedal locomotion was perhaps highlighted by five of the top eight robots using trundling transportation as well as walking. Rolling rather than walking offered greater stability in many situations.
Putting one foot in front of the other can be awesomely tricky in non-laboratory settings.
“Walking is not a form of artificial intelligence; it’s rigorous algorithms to not fall down,” explained one competitor.
And of all the many, many falls in the finals, only one saw a robot get back up: CMU’s third-placed Chimp.
Oh said human rescue robots, at this stage, “need to be rescued before they rescue others”.
A late starter
Oh’s passion for human-like robotics came late. After studying his bachelor and master’s degrees in mechanical engineering in Korea, he earned his PhD from University of California, Berkeley, in 1985, completing his thesis on automating milling tasks.Professor Oh and DRC-Hubo with Team KAIST at the DARPA Challenge in Los Angeles last year.
His subsequent work and research involved areas such as industrial operator control, manufacturing microprocessor applications, sensors and system integration.
“Walking is not a form of artificial intelligence; it’s rigorous algorithms to not fall down.”
Around 2000 he was inspired to build a robot by Honda’s Asimov, which he still refers to as a “beautiful human-like robot”. After a prototype involving a simple pair of legs built in 2001, Oh worked up to a proper human-like form factor – Korea’s first ever – in late 2004. This was the KHR-3 Hubo.
Following that success, he gained seven years’ grant funding of about $US 500,000 annually and then launched a spin-out company, Rainbow Robotics, to manufacture and sell Hubo robots. Around 25 have been delivered so far, including in the US, China and Singapore.
Autonomous robotics are not a focus at Hubo Lab. Rainbow sells Hubo as a platform with many motion primitives and an in-house real-time motion control system called ‘PODO’, which Oh believes helped with complicated tasks such as those in the DARPA Challenge, and had advantages over the popular robot operating system (ROS).DRC-Hubo exiting a vehicle during the DARPA Challenge.
“[ROS] is quite high-level structure, so to operate some of the motors and sensors there are many decorative agents in human-like robots – more than 30, 40, maybe 50 agents,” he said.
“It is very hard to operate them in real-time, at the same time all together.”
Where human-like robots are useful
Human-like robots are expensive to make, difficult to use and not yet useful enough to be put to work in general tasks – at least at the moment.
Recently, Alphabet announced it will sell Boston Dynamics, maker of the Atlas platform and weapon of choice for no fewer than six DARPA finalists. The reason given was that it’s unlikely to have a marketable product any time soon.
Robots have been a commercial success for simple, repetitive tasks such as vacuuming and pool cleaning. However, it’s difficult to see where sophisticated, multi-purpose human-like robots would be useful in anything but research at the moment.
The DRC Hubo costs an estimated US$500,000 to US$1 million to build. The development of robots at Hubo Labs has culminated in a machine that still generally needed people in the loop to perform any tasks. And this was on the way to winning arguably the most prestigious robotics challenge this decade.
This is not to insult the efforts of the champions of Team KAIST. DRC Hubo’s impressive egress after driving a car was autonomous. Apart from this, it was common to have at least some tele-operation involved in tasks. This ranged from indicating a ‘region of interest’ for the robot to target, or judging values from lidar information before indicating for Hubo to proceed (for example to climb steps). At the other end of the spectrum was driving a car, which was 100 per cent manually controlled.
Robots versus employment
Automation anxiety has been obvious in the past few years. A 2013 paper by Oxford Martin School scholars Dr Michael Osborne and Dr Carl Frey predicted nearly half of US jobs are at risk of automation in the next two decades, a statistic regularly cited in doom-and-gloom news articles.
There, as elsewhere, the prowess of robots and artificial intelligence – and the threat to jobs – has been exaggerated, Oh said.
“They might take some labour or some part of our jobs. What I’m saying is that a robot or AI can do a job that is usually dirty and difficult work [or] very habitual,” he said.
Advice to a young roboticist
For reasons such as the coming autonomous car era, companies are on the lookout for bright roboticists (see for example, Uber’s raid on CMU’s National Robotics Engineering Center last year).
Professor Jun-Ho Oh is often asked what advice he’d give to a young would-be robot maker, keen to get started in the exciting, in-demand professions.
First of all, if you want to commit to the field, only do so after you’ve had the chance to go to university for a while. The field is strongly multidisciplinary. Therefore, keep your interests broad. Also, be prepared for a lot of very hard work and immense demands on your creativity.
Robotics isn’t meant to be easy. But take courage, as the saying goes, it can delightful.
“I’m going to say, yes, play with robots,” the professor advises a youngster who wants to tinker.
“Play with anything! Keep your imagination, keep your curiosity, the interest on the general side of both science and technology, because robotics is not a single technology. It’s a production of the imagination.
“To become a roboticist, you need a very high level of imagination to power it.”