Imagine an earthquake or nuclear accident disaster area where it’s impossible to send in rescuers without putting their lives at risk. At UCL, Nicolas Van der Noot is looking to robots to do the job.
In science fiction films robots can move like humans, but that’s a far cry from the reality in robotics laboratories. ‘In most films robots are played by humans swaddled in sensors and computer-generated imagery’, explains Nicolas Van der Noot, a Belgian Scientific Research Fund doctoral student at the UCL Centre for Research in Mechatronics and at the Biorobotics Laboratory of the École Polytechnique Fédérale de Lausanne (EPFL). ‘Today’s industrial robots are incapable of moving like us. At best, they can walk on bent knees, which are neither sturdy enough nor energy efficient.’
On top of that, robots aren’t ready to leave the laboratory: they can only move over a flat surface free of obstacles, and they usually fail when ordered to walk over rugged terrain. ‘Developing robots that can mimic our walk is one of the great challenges of our time.’
Robots capable of human movement could not only reduce risk to rescuers but prevent disasters in the first place. Had someone—or something—been able to go inside the Fukushima nuclear reactor and close a few valves, huge quantities of radioactive vapour would not have escaped. ‘But the radioactivity was too high for a human’, Mr Van der Noot says. ‘Only a robot would have been able to do it.’ He’s trying to invent one that could.
Inspired by the human gait
To do so, he chose a gait inspired on one hand by traditional industrial robotics and on the other by human walking modelling experiments. ‘To us, walking seems quite simple, but it’s the result of millions of years of evolution whose stages are not all known to us. If we’re going to use the human gait as our inspiration, we have to formulate and test hypotheses.’ His research is a two-stage process: implement human muscle properties in robots, then learn how to control them.
Implementing and controlling virtual muscles
To equip robots with muscular properties, grafting real muscle is of course out of the question. ‘It’s virtual muscle that only exists through algorithms that run on the robot’s computer, which is in a way the robot’s brain. The movement of these virtual muscles can be mimicked with the help of 23 small motors in the robot’s structure.’ But to make them do what you want, you have to know the best way to stimulate them.
‘That’s the second part of my work. I have to find out how to stimulate the robot to walk by applying the right force underfoot, or to move in accordance with the tilt of its back. To do so, I use computer simulations to test stimulation combinations that will gradually determine which provide the best gait, then refine them to achieve the ideal movement combining optimal motions and adequate energy.’ It’s a bit like a child learns how to walk: he stands and falls, gets up and take a few steps before falling again, etc. The brain gradually corrects errors and acquires the right mechanisms.
COMAN, the little robot
When Mr Van der Noot gets interesting results, he tests them using the COMAN robot acquired by the École Polytechnique Fédérale de Lausanne. ‘This humanoid robot is the size of a five-year-old and has everything I need to test my walking algorithms.’ It works rather well: he recently helped COMAN take 50 steps on a treadmill, holding its hand so it wouldn’t fall. ‘That’s already a big step. Of course, I have to be realistic, this science is in its infancy, so it’s still too early to imagine COMAN in a disaster area, but it’s evolving.’