A team of engineers and computer scientists from Duke University has developed a new computer processor designed so that robots can master ‘motion planning’.
Motion planning – for example the ability to reach around or avoid obstacles to select an object – is not easy for robots. With their multi-jointed limbs, a task as simple as picking up an object in an environment that has not been pre-engineered for the robot requires time-consuming computation.
But not for much longer. The Duke University team has developed a specially-designed computer processor for motion planning that has the ability to plan up to 10,000 times faster than existing methods, while consuming a fraction of the power.
The processor works so fast that it can plan and operate in real time.
“When you think about a car assembly line, the entire environment is carefully controlled so that the robots can blindly repeat the same movements over and over again,” explains George Konidaris, assistant professor of computer science and electrical and computer engineering at Duke University.
“The car parts are in exactly the same place every time, and the robots are contained within cages so that humans don’t wander past. But if your robot is using motion planning in real time and a part is in a different place, or there’s some unexpected clutter, or a human walks by, it’ll do the right thing.”
It is this ability to deal with the ‘unexpected’ that makes the development so exciting.
Motion planning in robotics has been studied for 30 years, and recent advances have brought the time required for a robot to plan down to just a few seconds. Most of these existing approaches, however, rely on general purpose CPUs (central processing units) or quicker, but less power-efficient, graphic processing units (GPUs).
“While a general-purpose CPU is good at many tasks, it cannot compete with a processor specially designed for just a single task,” says Daniel Sorin, professor of electrical and computer engineering and computer science at Duke.
The new processor has been designed to perform the all-important task of collision detection, which is the most time-consuming aspect of motion planning.
“We streamlined our design and focused our hardware and power budgets on just the specific tasks that matter for motion planning,” Sorin explained.
So how exactly does the technology work?
The processor breaks down the operating space of a robot’s arm into thousands of 3D volumes known as voxels. The algorithm then determines whether or not an object is present in one of the voxels contained within pre-programmed motion paths.
The technology is able to check thousands of motion paths at once, and it then creates the shortest motion path available using the remaining ‘safe’ options.
“The state of the art prior to our work used high-performance, commodity graphics processors that consume 200 to 300 watts. And even then,” Konidaris noted, “it was taking hundreds of milliseconds, or even as much as a second, to find a plan.
“We’re at less than a millisecond, and less than 10 watts. Even if we weren’t faster, the power savings alone will add up in factories with thousands, or even millions, of robots.”
The team’s processor boasts a distinct advantage in terms of time not wasted and power saved.
It also opens up new ways to use motion planning, according to Konidaris.
“Previously, planning was done once per movement, because it was so slow, but now it is fast enough that it could be used as a component of a more complex planning algorithm, perhaps one that sequences several simpler motions or plans ahead to reason about the movement of several objects.”
This could open up a range of new opportunities for automation, as robots become more power-efficient and perform tasks faster.
“Real-time motion planning could really be a game-changer for robotics,” Konidaris concluded.