In structured environments, such as on manufacturing lines, robots are able to carry out pre-planned movements much faster than humans, but in unfamiliar environments it takes a lot of time for robots to plan movements that humans make almost without thinking. To give robots a speed boost, engineers at Duke University have developed a new processor that enables robots to perform motion planning 10,000 times faster than conventional methods.
When dealing with unstructured environments, robots have to perform an incredible amount of computation. Instead of quickly moving from point A to B, the robot has to tentatively select a possible motion, scan the area it will move through, and assess every cubic inch for a possible collision. Since this involves calculations that can number in the millions, it's no wonder that the stereotypical robot is marked by a succession of move, long pause, move, long pause sequences.
Sick of Ads?
More than 700 New Atlas Plus subscribers read our newsletter and website without ads.
Join them for just US$19 a year.More Information
The Duke team's approach involves the use of bespoke processors rather than relying on general purpose ones and is claimed to improve the speed of the robot's motion planning by a factor of 10,000 while reducing power consumption. The key to this is to essentially hard-wire the decision-making process into the chip rather than relying on software algorithms.
For motion planning, the Duke team divided the area that a robot's arm moves through into little match-box-like spaces called voxels. These are scanned by an array of cameras that plots any stationary or moving objects in the area. This information is fed into the specialized processor, which is made up of a many sets of logic circuits. Each set represents a voxel and asks whether the voxel is empty or has something in it.
Duke says that this is much faster than a general purpose processor because, instead of doing calculations for each voxel to see if there's a potential collision, the technology can check thousands of possible paths simultaneously. The result is that the robot isn't spending seconds or even minutes calculating a move, but can carry out the task of finding the safest and straightest path in real time while using up to 30 times less power.
"Previously, planning was done once per movement, because it was so slow," says George Konidaris, assistant professor of computer science and electrical and computer engineering. "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."
The team sees the new motion planning technology has having a large number of applications and a spinoff company called Realtime Robotic has been established to further develop it.
The robotic motion planning system will be presented at the 2016 Robotic Science and Systems conference at the University of Michigan.
The video below shows the robot motion planning system going through the, well, motions.Source: Duke University