Teaching a robot to fall over without making a fool of itself
Roboticists at the Georgia Institute of Technology are teaching robots how to brace themselves for a fall.
Humanoid robots are still, as we've seen from this year's DARPA Robotics Challenge, not exactly feasible. While the compilation video of robots falling over was worth a guffaw or two, it also demonstrated a fundamental flaw in robotics: lack of instincts and reflexes.
When a human trips and falls, they instinctively move their body to minimise damage to the most vulnerable parts of the body, twisting to protect those parts and putting out arms or legs to take the brunt of the impact. After all, a broken arm is fixable. A broken brain, not so much.
Because robots don't have these instincts and reflexes, they tend to fall flat on their faces. Granted, metal robots are a little sturdier than flesh-and-blood humans, but that doesn't mean they're immune to damage. So to help protect robots' more delicate parts, researchers at the Georgia Institute of Technology are choosing to sacrifice amusement.
In other words, they're teaching robots how to fall more like a human.
"A fall can potentially cause detrimental damage to the robot and enormous cost to repair," said Sehoon Ha, a PhD graduate who, along with professor Karen Liu, developed a new algorithm that tells robots how to respond to falls.
"We believe robots can learn how to fall safely. Our work unified existing research about how to teach robots to fall by giving them a tool to automatically determine the total number of contacts (how many hands shoved it, for example), the order of contacts, and the position and timing of those contacts. All of that impacts the potential of a fall and changes the robot's response."
The algorithm gives the robot tools to respond to a wide variety of falls, from a gentle trip or nudge to a rolling tumble. It also allows the robot to learn the best sequence of movements to slow a fall, as demonstrated in experimental tests on a BioLoid GP humanoid robot.
It is not just humans from which Ha and Liu drew inspiration. Liu had conducted previous research into how cats twist their bodies while falling in order to land on their feet. From this, she determined that the angle of impact was one of the most important factors in softening a fall. The challenge was teaching a robot how to use its rather stiffer body to achieve a similar result to a lithe cat.
"From previous work, we knew a robot had the computational know-how to achieve a softer landing, but it didn't have the hardware to move quickly enough like a cat," Liu said. "Our new planning algorithm takes into account the hardware constraints and the capabilities of the robot, and suggests a sequence of contacts so the robot gradually can slow itself down."
The results aren't, perhaps, hugely impressive by cat standards, but it's definitely a stumble in the right direction. So you had better make the most of enjoying this while you have the chance.
Ha and Liu presented their research at the IEEE/RSJ International Conference on Intelligent Robots and Systems in Hamburg, Germany, earlier this month.